[EM] Did someone not hear what I said about Approval vs Condorcet?

[Kristofer Munsterhjelm]

On 2024-05-26 07:28, Michael Ossipoff wrote:
> Someone keeps repeating that the voters shouldn’t have to vote 
> strategically. He wants the method to do it all for us, after we merely 
> state our sincere-rankings.
> 
> That’s of course a common attitude:
> 
> …wanting a high-tech, computation-intensive,computer-dependent system to 
> do it all for us, taking all the actual choosing responsibility off of 
> us.…sheltering & isolating us from the choice.

I would prefer that you do not attribute opinions that the proponents 
have not expressed. Nowhere have I said that Condorcet "[isolates] us 
from the choice" we make.

What you call "sheltering" and "isolating", I see as the method taking 
proper responsibility - proper responsibility to turn the voters' 
unambiguous honest opinions into an outcome without dumping the 
algorithmic calculation upon te voter themselves. See my "kick the can 
down the road" post for more info.

Now, I could make a caricature of Approval itself. Perhaps something 
about a calculator that just says "IDK, do the base conversion yourself, 
I only accept input numbers in factoradic". But caricatures only make 
people angry. Let's not stoop to them, shall we?

> I’ll ask this for the 3^rd time:
> 
> …
> 
> How would like you to handount-audit a Condorcet count for a 
> many-candidate national presidential election?

I think your question assumes something that won't hold. If you have a 
25-candidate presidential election, you've already lost, because nobody 
is going to rank 25 candidates, irrespective of whether the method is 
Condorcet, IRV, or Borda.

I'm not familiar with minor parties in the US. Has there ever been a 
25-candidate presidential election?

I don't think I can comment beyond that: I don't know enough about poll 
workers. I'll leave that to someone with experience. (Although roughly 
calculating: suppose 6 candidates like in Burlington. That's 30 pairs. 
Five times the work if counting a particular preference is as before. So 
you'd either need 5x the workers, or five times the time, or some 
combination of the two.)

On an aside, though, I would say that I generally wouldn't want to get 
computers anywhere near election counting. However, if you absolutely 
have to have them, there are ways of making sure they don't cheat: 
formal verification. You could also create special-purpose tools that 
say, only turn ranks into matrices and nothing else: there's no reason 
(apart from programmer convenience) why an election tool should be a 
general purpose computer that you could hide all sorts of shenanigans in.

> I’ve discussed that at length in previous posts, & it probably isn’t 
> necessary to again post about ways of choosing how to vote in Approval.
> 
> But, just summarize:It’s easy.Whichever of the various ways you prefer 
> to use, for choosing whom to approve, it’s easy.…& no, it doesn’t 
> require knowing your objectively-optimal vote.

> 
> I’ve many times pointed out that Approval’s Myerson-Weber equilibrium is 
> the voter-median.
> 
> i.e. Approval soon homes in on where the Condorcet-Winner is.

Not necessarily. See the following paper: https://arxiv.org/abs/2001.04216v2

According to the authors, under their model, if every voter follows a 
particular thresholding rule, then iterative approval generally arrives 
at the Condorcet winner (but not always). However, if voters are left to 
choose any thresholding rule they want (as you propose), then anything 
is possible. The arrangement might even elect a Condorcet loser, and the 
outcome may be slow to converge or who wins may depend on how many polls 
you hold.

I'm reminded of a quote about distributed algorithms that I read 
somewhere: "It's really easy to design distributed algorithms that 
suffer from deadlock, network floods or widely unpredictable and bizarre 
oscillations". That's in the context of computer science - deadlocks 
might not be applicable to election methods. But it does justify a 
starting position of skepticisim when considering schemes that offload 
more of the work to the voters by turning a one-shot method into a 
dynamical system.

> It seems to me that, in every one of EM’s polls, including the recent 
> one, Approval chose the CW.
> 
> Have we forgotten that?

I haven't, nor have I forgotten that Approval wasn't actually the poll 
winner.

But let's take this reasoning at face value. I don't have the other 
polls' ballot data available at the moment, so let's consider the latest 
one and pick... say, Borda.

As one can see by going to https://munsterhjelm.no/km/rbvote/calc.html, 
pasting in the data, and clicking Borda, Ranked Pairs is also the Borda 
winner.

But I don't think I'm going to start advocating for Borda.


To not be accused of tu quoque, let me clarify the point. As Burlington 
shows, an election method needs to handle the hard cases, not just the 
easy ones, or there may be a considerable uproar when the method is 
faced with a hard case and then stumbles. (In addition, people trying to 
make sure a stumble doesn't happen may start to do mass compromising, 
further entrenching two-party rule.)

So it's quite possible that our polls are easy cases. But like FairVote 
claiming that IRV gets the Condorcet winner more than 90% of the time, 
that doesn't by itself tell us much, because the failures have such a 
strong impact.

I illustrate the example above by picking a method we know to be bad 
(Borda being so extremely easy to fool with cloning and burial), and 
showing that the poll result comes out right. If a bad method can get a 
good result, then "getting a good result" is less useful than it might 
appear at first glance.

-km