[EM] Better Choices for Democracy

[Kristofer Munsterhjelm]

On 2025-06-24 21:48, robert bristow-johnson via Election-Methods wrote:
> 
> 
>> On 06/24/2025 2:34 PM EDT Kristofer Munsterhjelm <km-elmet at munsterhjelm.no> wrote:
>>
>> What do you mean by that "the law should say what it means and mean what
>> it says"?
> 
> This is how I interpreted it when this was said to me by my state
> rep  and the legislative counsel.  This is as pertains to this bill, (which
> went nowhere, unfortunately) in the previous legislature:
> https://legislature.vermont.gov/Documents/2024/Docs/BILLS/H-0424/H-0424%20As%20Introduced.pdf


> So then, of course, the question was *which* Condorcet method and
> how  will the actually legislative language be drafted?
> 
> We were looking at BTR-IRV as one possibility, simply because everyone
> knew about IRV and this was the simplest modification to IRV that people
> were familiar with.  There were two problems, we could still get
> pairwise tallies that are summable to see who the CW is, but the BTR-IRV
> method didn't rely on all of these tallies but *only* the pair tallies
> for the bottom-two candidates in each elimination round.  And, it seemed
> to the lawyer (the legislative counselor) and the state rep that this
> language was an "end-around" to sorta disguise or obscure the true
> purpose of the bill and that people will trust our intentions more if we
> just say clearly what it is we want to accomplish.
> 
> What we want is the Condorcet criterion to be adhered to whenever
> possible and when it cannot, the simplest and fair resolution to the
> election that most people, as a whole, would accept is fair and makes
> sense.  This basically told us that we needed to codify the
> straight-forward "two-method system" Condorcet language ...
> 
> Earlier in the bill they define what a "paired comparison" is.  We
> started to get a little hesitant to make it too complicated and we
> settled on Condorcet-Plurality.  But, if we do this again, I will try to
> persuade the sponsoring rep to do it a little better (Condorcet-TTR).
> 
> This is what we got to, to spell out to everyone in the room, in
> simple and plain English, what the proposed legislation is meant to
> accomplish.  This is what they meant when they told me that "the law
> should simply say what it means and mean what it says."  If it were
> Schulze or even Ranked Pairs, that language would be unclear to most
> people reading it what, in principle, the legislation is intended to
> accomplish.  (Sorry Markus, sorry Nic.)  Perhaps Minimax could suffice
> for legislative language.
> 
>> Consider a method like IRV. Procedurally, it is simple: repeatedly
>> eliminate the Plurality loser until someone has a majority.
> 
> Of course, sometimes that doesn't happen.  This is one of FairVote's
> lies.  48% is no majority.

Yes, I used IRV language for simplicity's sake. There are two devils in 
the details (majority of continuing voters vs all voters, and majority 
as Condorcet winner vs. IRV majority only ensuring it passes Condorcet 
loser).

>> ... But in
>> practice, it can exhibit chaotic behavior, nonmonotonicity, and so on.
>> How its procedure works doesn't seem to tell you much about how it behaves.
>>
>> So if IRV is the standard, then the required connection between what the
>> law says (the procedure it defines) and what it means (what happens as a
>> consequence) doesn't seem very strict, and it doesn't seem like BTR-IRV
>> would be any worse in that respect.
> 
> I agree with all of that, but I read that as a problem for IRV.  The
> technical thing that we want RCV to do for us is to prevent the Spoiler
> Effect by granting to voters that their 2nd-choice vote will be counted
> in lieu of their 1st-choice when their 1st-choice candidate is defeated.

Alright, so I read that as that what is being proposed should follow the 
spirit of two properties:

First, that the method should be clear about what it's using its 
pairwise data for, so that the voters can follow how it passes Condorcet 
(and thus ensures majority rule in the Condorcet winner sense), and it's 
not just "pairwise data goes into a mixer here".

And second, that what's being proposed should hold itself to a higher 
standard than IRV, and at least try to make it more clear how it 
behaves, not just what its procedure is.

Then it makes sense to do a composite method if the voters are already 
familiar with the base method, because it uses the Condorcet part just 
to determine if there's a Condorcet winner or not. Alternatively, to use 
something that's through-and-through pairwise while being simple enough 
- e.g minmax, an agenda method, or elimination tournament.[1]

My main concern with composite methods, particularly where the base 
method so thoroughly fails Condorcet as Plurality does, is that by 
chance or strategy, there is a cycle, and then the base method, that 
hasn't been used before, blows up in the voters' faces.

For C//Plurality, Plurality heavily punishes multipartyism. So you could 
have a great diversity of candidates, and then once in a while, there's 
a cycle (even if random), and then the result becomes very random 
because of Plurality's propensity to vote-splitting. Then it heals 
itself (if the method isn't repealed) and again supports many candidates 
until the next oops happens. C//TTR is somewhat better because TTR has a 
relatively low strategy rate (at least with not too many candidates), so 
there would be less of an incentive to blow things up on purpose.

But maybe that's just a tough deal forced by the constraints that the 
method should be legible and not use pairwise data in a strange way.

-km

[1] The requirement that the method is summable would also exclude every 
IRV-based method, even BTR-IRV, because you need the subelection 
plurality counts to determine who the loser/s are, and those aren't 
summable.