Election security under IRV

[rob]

@rb-j said in Election security under IRV:

Because 43% of the city marked their ballots that Andy Montroll was preferred to Bob while 36% marked their ballots that Bob was preferred to Andy.

One out of 440 RCV elections. Close 3 way race. Only 9000 voters, meaning quirks are more likely.

No, I don't prefer FPTP to that, and I honestly can't understand how you could. There's no indication Montroll would have won under FPTP. Kiss could have won under approval or STAR as well, we can't know.

But again, I think one minor failure in 440 elections is not even in the same ballpark with the downsides of FPTP. That's all I have to say on it.

[Jack Waugh]

Maybe some of you are talking past each other. Based on Rob's argument, I'm envisioning that the precincts publish the data from the cast ballots over the Internet where anybody can get those data and replicate the tally. Where's the single point of vulnerability if that is done?

[rb-j]

@rob said in Election security under IRV:

No, I don't prefer FPTP to that, and I honestly can't understand how you could.

I don't recall ever writing or saying that I preferred FPTP to any RCV. What I said is:

1. FPTP has precinct summability. Hare RCV does not. Condorcet RCV does.

2. Just because I prefer FPTP less than Hare RCV does not mean I should be satisfied with Hare RCV and I am not. We want RCV to (a) guarantee the majority candidate is elected, (b) eliminate the spoiler effect and (c) allow voters to vote non-tactically for the candidate they really want to win. When RCV failed to do that, not because of anything inherent to RCV but solely because of the method of tallying the ballots, we should act to fix these problems that RCV promises to fix. Denial of these problems (which is what FairVote does) is not conducive to fixing them.

There's no indication Montroll would have won under FPTP.

So what? Doesn't change anything regarding the objective failure of Hare RCV to elect the majority candidate and to eliminate the spoiler effect and to disincentivize tactical voting (by punishing 1/6th of the electorate for "voting their hopes rather than their fears", they would have been better off voting their fears).

But again, I think one minor failure in 440 elections is not even in the same ballpark with the downsides of FPTP. That's all I have to say on it.

While this happened only once (that we know of in the United States), it can happen again.

Occasionally we read about surgeons accidentally amputating the wrong limb because of a procedural mixup (just a "minor failure"). When this happens, do we hear hospital officials defending their procedures saying “These procedures have served us well for decades and hundreds of surgeries, so we see no need to change our procedure at all.”?

Even with a single catastrophic mistake, that is enough to motivate review and make changes in procedure to insure that such an unnecessary failure will not happen again. Now is precisely the time to recognize and correct this failure that occurred in Burlington in 2009 rather than sweeping it under the rug and hoping no one notices.

[rb-j]

@jack-waugh said in Election security under IRV:

Based on Rob's argument, I'm envisioning that the precincts publish the data from the cast ballots over the Internet where anybody can get those data and replicate the tally.

The data for Hare RCV is too much for anyone simply "replicate the tally". You're missing the whole point of precinct summability and what it does for process transparency.

Again, imagine, just with FPTP, if the government passes a law that proscribes polling places from publishing the vote tallies of each candidate in the FPTP race. (Imagine they publish the data from each cast ballot on the internet, instead.) Now what are media organizations or the competing campaigns gonna do on election night to verify the results? Especially if results are not announced until four days after the election?

[Jack Waugh]

@rb-j said in Election security under IRV:

The data for Hare RCV is too much for anyone simply "replicate the tally".

Anyone with an Internet-connected computer can replicate it. I'm proposing that all the precincts do so as a matter of course. They would all be using the same software, and so should publish identical results unless something is seriously wrong.

(Imagine they publish the data from each cast ballot on the internet, instead.) Now what are media organizations or the competing campaigns gonna do on election night to verify the results?

They are going to run standard computer programs that suck the cast-ballot data across the Internet and do the tally.

[A Former User]

@rb-j said in Election security under IRV:

Now what are media organizations or the competing campaigns gonna do on election night to verify the results? Especially if results are not announced until four days after the election?

import pandas as pd
from usaelections import irv

df = pd.read_csv("ballots_from_my_awesome_precinct.csv")

projected_winner = "Boe Jiden"
actual_winner = irv(df)

if not actual_winner == projected_winner:
    call_the_FBI_immediately()

[rob]

@rb-j said in Election security under IRV:

I don't recall ever writing or saying that I preferred FPTP to any RCV.

I can't claim to know your position on that, all I know is what I I've seen you write online.

And I think the net effect of what you've said, especially if lots of people took your words to heart, is simply to reduce the chance of there being any reform at all. You seem to have put far more words and energy into tearing down ranked-choice, versus advancing anything that actually has a chance of making an impact in our lifetimes.

That's just my observation. I believe I am in your camp on advocating for Condorcet methods. I just think we have different priorities and approaches. For instance, precinct summability, while a positive, is down near the bottom of my list of important things. In terms of getting things implemented, far more important is "has been extensively tested," and "there is existing legislation we can draw from" and "we can use existing voting machines and infrastructure" --- and RCV-Hare has that over any Condorcet method (as well as Score, Approval, etc). Not to mention a large organizations with some funding (Fairvote, Forward Party), public mindshare, etc.

I do think that bottom two runoff, which I learned about from you (thanks!) is potentially a way to "sneak" a better system in. It just seems so similar to RCV-Hare that the RCV people are likely to say "whatever, that's fine too."

But the key is to that is to build up RCV-BTR without spreading fear uncertainly and doubt about RCV-Hare.

[Jack Waugh]

@rob "we can use existing voting machines and infrastructure"

I already addressed this for Score with small numbers of options (7, not 100).

[rob]

@rb-j said in Election security under IRV:

The data for Hare RCV is too much for anyone simply "replicate the tally".

No it isn't. You can do it as simply as putting it into a Codepen and sharing it with the world, letting anyone fork it, make visualizations, etc.

Here is a visualization someone did based on ballot data from a Hare-RCV election. (it wasn't done in a CodePen, but that's something I am working on)

Even for a state-wide election, the actual ballot data can be pretty small, if you just compress it into a format like I've been using, which doesn't list duplicate ballots multiple times, it just says the number of ballots prior to each line.

I would expect precincts could deliver all their results that way in an email. (with official results delivered another way, of course)

e.g.:

a: Mark Leno
b: London Breed
c: Jane Kim
d: Angela Alioto

203453: b>c>d>a
193234: b>d>c>a
123934: b>d>c

(and so on)

[rb-j]

@jack-waugh said in Election security under IRV:

Anyone with an Internet-connected computer can replicate it.

Opaquely.

I'm proposing that all the precincts do so as a matter of course.

It's still opaque.

They would all be using the same software, and so should publish identical results unless something is seriously wrong.

Please be realistic. What can you expect media organizations and competing campaigns to do?

1. Run opaque software on raw data and accept the winner that pops up.

or

2. add N(N-1) tallies up (it's 20 tallies using Condorcet or 5 tallies using FPTP for five candidates) and see who beats who.

One of these processes is transparent. The other is not.

For election security that stands beyond suspicion, we want every step in the process to be transparent. Where ordinary people can see, understand, believe, and accept the results.

[rob]

@jack-waugh said in Election security under IRV:

we can use existing voting machines and infrastructure"

I already addressed this for Score with small numbers of options (7, not 100).

You may be right. However, I saw this thread, where choco_pi says otherwise. Based on all the other stuff he has written, and the detail in his answer, I'm inclined to believe he knows what he is talking about.

https://www.reddit.com/r/EndFPTP/comments/w1r6f4/why_is_score_voting_controversial_in_this_sub/

But it's said that current machines can handle score voting. Do you have proof on your statement?

It's hard to respond to claims like this because they're just, super wrong. Like it's so disconnected with the facts on the ground it's difficult to know where to begin.

Voting systems are not just a single machine, but the entire chain of hardware, software, instructions, and protocols covering everything from ballot creation, vote casting, ballot scanning, adjucation, reporting, verification, and auditing. For federal and state elections, the entire system A-Z has to be certified according to law, including equipment testing at designated labs. This includes not just security, but a litany of accessibility concerns.

There are only 11 organizations even registered with the EAC as qualified to even submit a system for certification. (Of these, Vidaloop is just a startup seemingly focused on mobile voting, so it is really 10.)

3 of these, ES&S, Dominion, and Hart, make up over 92% of the US market. All have end-to-end RCV ballot support on their current generation machines. Unisyn and Avante do as well.

The software for Clear Ballot and various legacy systems by the aforementioned vendors (such as Sequoia, who I believe Dominion purchased) do not support ranked ballot tabulation directly, but export ballot-level results that can still be tabulated externally. To plug this gap, RCVRC has invested a lot of resources into supplying an open source version of this missing piece and getting it certified in multiple states. (So that one Clear Ballot machine in Sheboygan Country doesn't block the entire state of Wisconsin) Smartmatic and VotingWorks are probably in a similar boat, but it has not yet come up that such an implementation has needed certification. (VotingWorks is itself open-source, so already certified functionality could just be integrated directly anyhow.)

Conversely, 0 of these 11 vendors support scored ballots. Not the ballots, not the voting machine UI, not the tabulation, not the reporting, not the auditing; none of it. You could try do an externally tabulated workaround, but in this case I think you'd find the reality of that compromise to be unacceptably ugly, 10x worse than the RCVTab stopgap. (A ballot with a question asking which candidates you would like to score 0, a seperate question asking which candidates you would score 1, etc.)

Approval sneaks in because you can take frequently take a plurality systems and just turn off overvoting detection. You still have a lot of work to do--interface software has to support it, accessibility instructions have to still be valid, LEO instructions need to be updated--but this is a realistic, immediately achievable set of changes if state law allows it.

Warren Smith yelling that there is no reason vendors can't make voting machines that support score ballots is like yelling that there's no reason car manufacturers can't make cars that run on vegetable oil. I mean... sure. But, they don't, which is the topic at hand.

[rb-j]

@rob said in Election security under IRV:

Warren Smith yelling that there is no reason vendors can't make voting machines that support score ballots is like yelling that there's no reason car manufacturers can't make cars that run on vegetable oil. I mean... sure. But, they don't, which is the topic at hand.

Zero vendors have software that does Condorcet. Doesn't mean that it can't be done, and if a state legislature adopts Condorcet, I am quite sure that Dominion will quickly do it to the state's specification. The Imagecast Precinct tabulator machines will have to have their firmware updated (so that pairwise defeat tallies are printed, which makes it precinct summable) and the Democracy Suite software will have another RCV Method added to the menu.

[rob]

@rb-j said in Election security under IRV:

It's still opaque.

I'm being totally transparent with the work I am doing. It will allow anyone to replicate it. If they want to look at the code, they look at the code. If they want to change the code, they can change the code. If they want to re-fetch the ballot data from the source and paste it in, they can do that. If they want to copy the code and run it on their own website, they can do that.

It all runs on a Codepen that is trivial to fork and modify, and you don't need to download a thing because it is just Javascript running in a browser.

And the code isn't complicated at all. I'd certainly welcome anyone with coding chops to dive in and help it be better, cleaner, etc. My wheelhouse is graphics and interactivity, but just doing raw tabulation with text output is really, really easy.

I really don't see the problem here.

[rb-j]

@rob said in Election security under IRV:

precinct summability, while a positive, is down near the bottom of my list of important things.

Well, I'm glad this guy bumps it up a little farther from the bottom.

In terms of getting things implemented, far more important is "has been extensively tested," and "there is existing legislation we can draw from" and "we can use existing voting machines and infrastructure"

Yeah, I hear that all the time and it's zero justification for violating majority rule (and one-person-one-vote), demonstrating the spoiler effect, punishing voters for voting sincerely, incentivizing tactical voting, and losing process transparency.

Also, remember this election? Can you imagine a large city or statewide RCV election in which Combined Write-In wins? With Hare, it's an awful mess. With precinct summability, it's far simpler and more secure.

"has been extensively tested,"

And the result of that "extensive testing" is that when the Condorcet winner is elected, people generally accept the results and the election is a success. When the "extensive testing" examines real-world elections in which the Condorcet winner exists and is not elected, only failure results.

Despite FairVote and VPIRG touting that Ranked-Choice Voting is “well tested”, the history shows that as long as the Hare RCV method elects the Condorcet winner, people generally accept the election outcome. But when Hare RCV does not elect the Condorcet winner, the election is considered a failure in every respect. Burlington 2009 demonstrates this perfectly as RCV was repealed the following year. A wide majority of the city did not see Bob Kiss as representing the majority of the city. We were not protected from the spoiler effect and a large portion of the city were punished simply for voting their hopes rather than their fears.

[rob]

@rb-j said in Election security under IRV:

Zero vendors have software that does Condorcet. Doesn't mean that it can't be done, and if a state legislature adopts Condorcet, I am quite sure that Dominion will quickly do it to the state's specification.

(to be clear, the quoted text wasn't my words, and it was about Cardinal ballots, not ranked ballots as could be used for Condorcet)

I don't doubt that, but you can't expect it to start with a state legislature, it will start with a smaller locality.

Regardless it is a chicken and egg problem. Legislatures are orders of magnitude less likely to adopt things that aren't already proven in practice. I think we can get there (i.e. condorcet, regardless of ballot type), but need to do it one step at a time. RCV first, then better RCV, is a very reasonable plan.

In the case of my home town (San Francisco), they started with RCV Hare but with only 3 rankings. It took many years before we got up to a dozen (but the legislation was written such that they could upgrade it at the discretion of the election official, rather than having to change the legislation) To me that is the exact "baby steps" approach that can work.

If a city is open to Condorcet, by all means, go for it. If they'd rather just go with something that is identical to something somewhere else, I say just let them do something they are more comfortable with so progress is made. I think it would be awesome to try to convince San Francisco, a city thoroughly comfortable with ranked ballots, to make it Condorcet. But you can do that without risking scaring people off from even moving away from plurality/choose one.

[rob]

@rb-j said in Election security under IRV:

Yeah, I hear that all the time and it's zero justification for violating majority rule (and one-person-one-vote), demonstrating the spoiler effect, punishing voters for voting sincerely, incentivizing tactical voting, and losing process transparency.

I'm very confused about your idea of majority rule / one-person-one-vote. You've said clearly that "majority rule" doesn't even make sense in a preferential ballot situation. I don't even know what "one person one vote" means.... to me, casting one ballot is one vote. RCV has been challenged on that basis and passed.

But I don't see how Condorcet methods magically get around that.

Punishing voters for voting sincerely and incentivizing tactical voting? Can you even show me an example of an attempt at that under Hare RCV? If it was realistically possible, I'm sure someone on Twitter would telling people to vote in a particular way to game an IRV vote. I've never seen it. Are they just whispering it among themselves offline?

Sorry, I'm not buying it.

losing process transparency.

Again, it's transparent enough for me but I'm not going to go down any more conspiracy theory rabbit holes. And I am working to make it more transparent by making easy ways people can re-run the tabulations or analyze things in incredibly open, trouble free and sharable ways. You're a coder, right? Why don't you help with that effort?

[rb-j]

@rob said in Election security under IRV:

I don't doubt that, but you can't expect it to start with a state legislature, it will start with a smaller locality.

Depends on how big your state is.

Regardless it is a chicken and egg problem. Legislatures are orders of magnitude less likely to adopt things that aren't already proven in practice.

Yes, and they pay attention when some thing demonstrates failure in practice.

I think we can get there (i.e. condorcet, regardless of ballot type), but need to do it one step at a time.

What steps precede the step of going from Hare to Condorcet?

RCV first, then better RCV, is a very reasonable plan.

Yah. We had IRV first. Then it fucked up. Then it got repealed. Now it's readopted in Burlington without acknowledging anything about 2009. And without that small bit of introspection, nothing will get fixed. Denial is not reform.

In the case of my home town (San Francisco), they started with RCV Hare but with only 3 rankings. It took many years before we got up to a dozen (but the legislation was written such that they could upgrade it at the discretion of the election official, rather than having to change the legislation) To me that is the exact "baby steps" approach that can work.

When you had only 3 rankings (and a dozen candidates), this was pointed to by RCV opponents as "voter disenfranchisement". And, in a sense, these RCV opponents were correct in that claim. But you would never have gotten that fixed if the RCV proponents simply ignored that problem as anything to worry about.

[rob]

@rb-j said in Election security under IRV:

Yah. We had IRV first. Then it fucked up. Then it got repealed. Now it's readopted in Burlington without acknowledging anything about 2009. And without that small bit of introspection, nothing will get fixed. Denial is not reform.

Your idea of "fucked up" is different from mine.

And it got repealed mostly because Wright and his voters were pissed off that he didn't win. But if Wright had won, that would have been a bigger fuck up, as I'm sure you know.

[rb-j]

@rob said in Election security under IRV:

I'm very confused about your idea of majority rule / one-person-one-vote.

Well, this is why it's good to read the paper. It's not just me.

You've said clearly that "majority rule" doesn't even make sense in a preferential ballot situation.

What, exactly, did I say clearly? I don't remember ever saying that "majority rule doesn't even make sense in a preferential ballot situation."

I don't even know what "one person one vote" means.... to me, casting one ballot is one vote.

So, if the vote represented on your ballot counted more (had more "punch") than the vote represented on my ballot and we're both a single person?

In 2000, 48.4% of American voters marked their ballots that Al Gore was preferred over George W. Bush while 47.9% marked their ballots to the contrary. Yet George W. Bush was elected to office. Did the greater number of voters preferring Gore have votes that counted as much as those from the fewer number of voters preferring Bush?

In 2016, 48.2% of American voters marked their ballots that Hillary Clinton was preferred over Donald Trump while 46.1% marked their ballots to the contrary. Yet Donald Trump was elected to office. Did the greater number of voters preferring Clinton have votes that counted as much as those from the fewer number of voters preferring Trump?

In 2009, 45.2% of Burlington voters marked their ballots tat Andy Montroll was preferred over Bob Kiss while 38.7% marked their ballots to the contrary. Yet Bob Kiss was elected to office. Did the greater number of voters preferring Montroll have votes that counted as much as those from the fewer number of voters preferring Kiss?

RCV has been challenged on that basis and passed.

... and repealed and passed again.

But I don't see how Condorcet methods magically get around that.

Of course there's Arrow and Gibbard and Gibbard–Satterthwaite, so even with Condorcet there can be circumstances where you simply cannot achieve majority rule without a contradiction. But that's when a Condorcet winner does not exist, a cycle.

When the Condorcet winner exists (so far, it looks pretty close to 100% of the time ranked ballots are used) and the Condorcet winner is elected, then majority rule is achieved and there is no sense that some voters' votes counted more than others.

Punishing voters for voting sincerely and incentivizing tactical voting? Can you even show me an example of an attempt at that under Hare RCV?

Remember there is a little difference between tactical voting (like compromising) and strategic voting (like burying). Incentivizing tactical voting happens when voters are punished for voting sincerely. That happened in Burlington in 2009. It always happens when an election is spoiled.

Violating majority rule and one-person-one-vote (they sorta go together) leads to a cascade of failures. It leads to a spoiled election and that leads to punishing some group of voters for voting sincerely and that incentivizes tactical voting which leads to tactical voting (the tactic being compromizing) in future elections.

If it was realistically possible, I'm sure someone on Twitter would telling people to vote in a particular way to game an IRV vote. I've never seen it.

That's strategic voting. I have never seen an attempt at burial (which doesn't work with Later-No-Harm, but would work with Borda). But I did see a legitimate strategic voting where two competing candidates endorsed each other publically. "Vote for me first, but rank this particular opponent of mine second."

Are they just whispering it among themselves offline?

With FPTP, people often whisper among themselves to "don't throw away your vote" and to compromize and vote for the major-party candidate best situated to beat the major-party candidate one loaths. This happens in elections that follow elections that were perceived as spoiled.

Now IRV in Burlington did demonstrate the spoiler effect in 2009 and people who voted for Kurt Wright found out that they actually caused the election of the candidate they disliked the most, simply by voting for the candidate they sincerely liked. How does that affect them in the next election?

Sorry, I'm not buying it.

Fortunately, you're not in the Vermont legislature nor Secretary of State's office (nor running for either).

losing process transparency.

Again, it's transparent enough for me but I'm not going to go down any more conspiracy theory rabbit holes. And I am working to make it more transparent by making easy ways people can re-run the tabulations

That's opaque. It's not transparent.

or analyze things in incredibly open, trouble free and sharable ways.

We cannot expect schlubs to read and understand some C++ code that someone puts on github. This is not transparent to them.

You're a coder, right? Why don't you help with that effort?

Last year I told Nicolaus Tideman I would try to write C code implementing Ranked-Pairs (using margins). Haven't done it yet. Someone somewhere has implemented Schulze, which is quite complicated.

[rb-j]

@rob said in Election security under IRV:

Your idea of "fucked up" is different from mine.

My idea of "fucked up" is

1. failing to elect the majority candidate after "guaranteeing a majority candidate is elected".

2. failing to protect against the spoiler effect after promising that "IRV eliminates the spoiler effect"

3. punishing voters who mark their favorite candidate as #1 after assuring everyone that they can "vote their hopes, not their fears".

And it got repealed mostly because Wright and his voters were pissed off that he didn't win.

There was a lotta misinformation flowing around. Especially from the GOP. I know. I was here. Considering post 2016, does that surprize you?

There was certainly a sense that Dems were robbed and that was spelled out at the time. Problem is that the Dems (including me) did not want to repeal IRV, but we had trouble defending it. The GOP just went hard with their "we were robbed" and "elect the candidate with the most votes" drivel. It still exists today (see VT guv's statement on H.744).

But if Wright had won, that would have been a bigger fuck up, as I'm sure you know.

With ranked ballots, we would know. Without ranked ballots it would appear that Wright had more support than anyone else. We had other spoiled elections in Vermont.

BTW, in my research on Burlington, I found these words from you, that I do agree with, am I correct that you've changed your mind in the years since? (or is there some other misunderstand I have about your position?)

You want simple? Here are the simple facts:1) A preference voting system like IRV is by far a much fairer method to use than the "simple" winner-take-all plurality system in the case when there are more than two viable candidates;2) This means that more voters were happy and/or comfortable with the prospect of Kiss as mayor than they would have been with any of the other candidates assuming the post;3) This is because IRV corrects the biggest flaw and frustration of the old plurality system: the spoiler problem. Hey, if you still don't get it, then just vote the old way and don't specify your second, third, etc. preferences. Under the IRV system, you still retain the right to only vote for your first choice. But don't you dare pretend that forcing others to vote their fears rather than their hopes is not a "problem." With all due respect, "Webber," what don't you get about any of the above?

I'd like to know where that came from. I do not use "winner-take-all" as a term for FPTP. Any single-winner election is "winner-take-all". I don't think those words are mine. But I yield to evidence. Maybe I was quoting someone else??