Rank choice voting unworkable for presidential elections


“I love how wonderfully vague your posts are. What specifically did I say that applies to all methods? How specifically does asymmetric encryption apply here?”

" But if there is any way for the fact that someone voted E > H > I > R > V > B > A > D > U > S > X > Y > Z to get exposed to the public, then we could have people bribe or threaten that you must vote exactly that way or else you and your family will be hunted down and destroyed.
You say anonymize the ballots. But then you figure out that H won, but you have to leak the decryption key to the public or else people may fear that someone in the ballot counting process artificially made “H” actually equal Dr. Fascist instead of Ima Goodperson, the person who was supposed to be labeled H.

My point is, you cannot have a transparent, secure, and bribe-free election under IRV although you can with Score Voting."

^ ALL of that


I know what asymmetric key cryptography is. I do not know how it applies to IRV elections in such a way that it dodges the problems I mentioned (anti bribery is the only one left standing, but it still does not appear to solve this while remaining transparent).


“I know what asymmetric key cryptography is”

So you were being disingenuous?!? You were deliberately making a dishonest argument?

No, just no. don’t do that. It’s never going to work and it’s a disrespectful waste of my time.


your nonsense FUD is just getting annoying. i’m going to stop feeding it.


With Score Voting each precinct can send a summary I.e. List of scores. You cannot reconstruct the individual votes but you can figure out the national winner, like so:
A: 2544
B: 5772
C: 636
…and so on. These would be independently verified and published widely. The most you could figure out is if a given candidate is voted on or not. You cannot get the kind of information you can in fully transparent IRV.
With IRV the precinct totals would look like this:
3 E
and so on. Ah, but I never saw the specific ordering EDBAJCMILKFGH, so that guy I tried to bribe failed! I guess no money for him…

I am not lying. You have not convinced me that asymmetric key cryptography can simultaneously provide a reasonably transparent election without allowing for bribery.


How about you provide a miniature example of how you run proposed system would work, from start to finish?


IRV does not eliminate the spoiler effect. So you would need the primary. Bernie could have prevented Clinton from winning without winning himself. Suppose the rank-order ballots look like:
27%: Sanders > Clinton > Trump
21%: Clinton > Sanders > Trump
4%: Clinton > Trump > Sanders
48%: Trump > {irrelevant}
If Bernie doesn’t run, Clinton defeats Trump 52-48.
If he does run, the first count is
Bernie: 27%
Clinton: 25%
So Clinton is eliminated, and Trump defeats Sanders 52-48 in the runoff.


In your example, you included 4% of Clinton voters putting Trump over Sanders. Given the dynamics of the election, this would have been unlikely as a sincere ranking, and looks more like a burial/Chicken-Dilemma strategy on the part of some of Clinton’s voters.

Using a Condorcet method that does not reward burial, such as MinLV(erw)SME (see https://github.com/dodecatheon/sorted-margins-elimination.git), this leads to a Trump victory.

If instead the 4% Clinton>Trump presumed burying voters switch to Clinton>Sanders, they end up losing to Sanders but prevent a Trump victory.

As you showed, IRV, which also does not reward burial, would also lead to a Trump victory. So there would be a strong disincentive for voters in an IRV race to attempt to bury their closest similar rivals.


You are so bad. 4% of the electorate is 16% of clinton voters, and even then that means 84% of the clinton voters went C>B>S… more than a 5:1 ratio. There will nearly always be exceptions. People might think Clinton & Trump are acceptable while Sanders is horrible.

But the broader concern is that you went after a minor quibble and blatantly ignored the larger concern that IRV does not prevent the Spoiler Effect.

Here I have a completely abstract IRV pathology example.
133 A>B>C
77 A>C>B
44 A
204 B>A>C
37 B>C>A
57 B
215 C>A>B
115 C>B>A

(This was, uh… totally randomly generated by my Election Pathology Generator Program.)

C is a spoiler and if about 70-80 CAB voters had refused to vote then it would have allowed the preferred A to win instead of B.


I could imagine (as an example) Orange County types voting Clinton > Trump > Sanders. Clinton defeated Trump there 609,961 votes to 507,148, but given that Orange County is generally conservative, I doubt Sanders would have done as well. Probably quite a few of them were willing to hold their nose for Clinton but would’ve found Sanders too far to the left.


You don’t need 40 candidates to ‘sign’ your ballot in a unique way so you can sell your vote. It takes just 12 candidates for the number of ways to fill out a ballot to exceed the population of the US. Furthermore, a large precinct might record ~3000 votes on a presidential election night. To exceed 3000 ways to fill out a ballot, you need 8 candidates (which is pretty typical: a google image search of presidential election ballots consistently shows around 7 or 8 candidates plus the write in slot.) That’s assuming that weakening the spoiler effect doesn’t encourage more people to try to get ballot access.
And of those 3000 voters, how many are going to bother to rank every Tom, Jill, and Gary who gets on the ballot? My estimate is that if you vote in a large precinct, then you’ll need to use 6 candidates plus the write-in slot* to sign a ballot so you can sell your vote.

*Assuming that write-ins will just be listed as WRITE IN on the reported rank-orders, and that no effort will be made to interpret who was written in.


Clearly the individual advocating RCV here is flagrantly misusing terminology. What we really mean by “precinct summable” is that the votes can simply be added up, announced, and forwarded to larger pools. This is rather simple. This is what happens with score/range voting, but not with any form of RCV. Rather, with any RCV, the ballots cannot ever be easily tallied at the precinct level, and then simply forwarded to larger pools one single time.

With RCV, the ballots themselves, or at least every bit of the data they embody, must be sent, either directly or via larger pools, up to some central authority which can either do the the final tallying and then decide which candidate is to be excluded, and then do another tally. and so on. Alternatively, they can do an initial tally to determine the candidate to be excluded, and send that name back to the precincts to be re-tallied

Then the entire process begins again, the precincts would send back the new tally, and yet another candidate would be excluded. If there are 100 candidates, this back-and-forth would require 99 cycles. Of course this would create a gigantic chain-of-custody problem.

Anyway, whenever RCV is attempted, history indicates that two-party capture emerges as surely as it does with the far more common single selection method. Also, nearly all technologically advanced nations use hand counted paper ballots, not automation. The devil lives in the details, and the computer is the epitome of limitless artificial complexity. And as the existence of gerrymandering attests, politicians are born hackers.

RCV/IRV is a malignant virus.


I just want to clarify that I am not trying to defend IRV generally, but there are some aspects of IRV that its defenders hold dear, and it doesn’t persuade anyone if you ignore those completely.

IRV is clone-independent, so my counter-example was to demonstrate that when Clinton and Sanders are within the same party, another clone-independent method, for example, the cited Condercet scheme, does not reward burying.

Incidentally, my Sorted Margins Elimination code on your second example yields A as Condorcet winner with 215 C>A>B voters, and B as Condorcet winner with 140 CAB voters. So the problem you are trying to illustrate with IRV is also an issue for Condorcet methods.

I happen to think that Condercet Methods have a lot going for them, so I’m just struggling to see what you’re after here.


I do not like Condorcet methods because they lead to other flaws, but in the example I gave there was a very clear and well-liked Condorcet winner, not a polarizing one that oppresses the (50-ε)% for the pleasure of the (50+ε)%.

I support Score Voting and STAR voting and 3-2-1 voting and Asset Voting. Two of those are clone-immune and AFAIK the others only reward, not punish, clones (but not like how you can guarantee a win by running hundreds of clones in Borda). Not sure about 321.

In real life there are no such things as perfect clones unless the electorate is very small. And that is the problem with IRV – it may be immune to perfect clones but if even a few voters break pattern we get spoilers again…




Both 3-2-1 and STAR fail clone independence. Clones won’t weaken a winner, but they may give one party an unfair advantage.

Also, if there are factions within a party, running clones can similarly create an unfair advantage there also.

And again, I am not arguing for IRV. I’ve simply been pointing out that in both your examples, you found the Condorcet winner unsatisfactory. But in some sense, both STAR and 3-2-1 are attempts to gain some of the pairwise effects of Condorcet, so that needs to be considered.


Yes and by that definition, RCV is “precinct summable”.


Wait. The problem with IRV is that such communication has to happen in two directions.
Additive Systems (choose1, Approval, borda, etc):

  1. Each precinct adds their votes and publishes their total.
  2. The precinct totals are added by the GCETS and the press and the winners are announced.


  1. Each precinct adds their votes and publishes the total.
  2. The precinct totals are added by the GCETS and the press, and either the winners are announced or the candidates to eliminate are announced.
  3. If there was no winner, the precincts start over with step 1.

This takes much longer and if you ever tried to e.g. buy a movie ticket when the line was long and there was only one ticket booth and then that one guy has some weird problem and holds up the entire line and… you know what this can be like.


there seems to be a lot of back and forth about the meaning of a phrase. but the meaning of a phrase isn’t what’s important. what’s important is what actually happens in reality or can happen. so let’s talk in those terms.

In a anked choice ballot, each election - each government position up for dispute, is independant. Additionally each possible way to fill out a ballot for a single government position in dispute, all such ballots can be totalledthis way. Two ballote that ranks candidates for a position in the order b,c,a can be added together and doubted simply as “2” ballota with that order. That is to say that at this points the ballots are “fungible”.

Such ballots can be totaled up at the precinct level. "946 ballots are a,c,d. 18753 are e,c,d,a,b. etc. " at this point each unique ordering follows the algebraic rules of both commutivity and associativity. you can add then up in any order the total is the same. you can total it up by county, and then by state. or you can skip county level and total it up by state. the total is the same.

grouped by ordering, for each individual election, tallying of the ballots is commutative and associative.


Gee that sounds kind of complicated. Why can’t you just add all the votes up one time and be done with it, like with simple score voting?

But the only really important thing is that RCV totally fails the elite party capture criterion, while SSV passes easily.


What about write in candidates?