[EM] The worst about each system; Approval Preferential Voting (new name for an MCA-like system)

[robert bristow-johnson]

i think it was Dave Ketchum who said:

> -Plurality: Everything. It routinely requires dishonest strategy  
> from a large minority, or even a majority, of voters. Enough said.

to which i answered:

> On Wed, May 26, 2010 at 11:30, robert bristow-johnson <rbj at audioimagination.com 
> > wrote:
> except some unnamed folks here
> (whose posts i don't see anymore) think that it's better than IRV.
>

On Jun 5, 2010, at 3:00 PM, clay shentrup wrote:

> Robert,
>
> I'm not necessarily saying that IRV is worse than plurality in terms  
> of direct election outcome performance.

Clay, you weren't that particular unnamed folk.  before i ventured  
over to the ESF forum, i had a little slugfest with Kathy Dopp (i  
think you know who she is).  even though she was not a direct  
participant, she *did* insert herself into a referendum we had here in  
Burlington Vermont about IRV (so did Warren, BTW, but Warren actually  
has a pretty good idea about the facts where Kathy was most just  
repeating partisan tripe).  the specific referendum was to replace IRV  
with the method that existed before, which is plurality if above 40%  
and top-two runoff (in 3 weeks) if no one gets 40%.  even though i  
have been clearly critical of IRV, particularly how well it performed  
(even regarding the very goals we had in adopting IRV in 2005) in  
2009, i voted against the recall.  but Kathy was clearly in favor of  
plurality over IRV.


> If you look at Bayesian Regret figures, IRV certainly is better. But  
> I have two major reasons for believing that, on the whole, it may be  
> worse than Plurality.
> 	• If enough IRV voters are strategic, then that improvement may be  
> too small to outweigh the side effects of IRV, such as
> 		• Greater risk of near-tie election recount nightmares (http://scorevoting.net/TieRisk.html 
> )

not if the ballot scanners are working.  if your ballot scanning  
technology is faulty, it's bad for any system.  there would be a lot  
more work in hand counting (and recounting after transferred votes) an  
IRV election.

> 		• Greater ballot spoilage rate (http://scorevoting.net/SPRates.html)

about 1/100 % in Burlington in 2009.  machine errors rejected 4 of  
7980 ballots, but upon examination 3 of those were okay.  only 1  
ballot in the entire city was a bad ballot in 2009.  i don't think the  
argument washes.

> 		• Need for central tabulation (http://scorevoting.net/ 
> IrvNonAdd.html)

to which i agree.  not so bad for a smaller venue, but not good for a  
state-wide or nationwide election.

> 		• Increased likelihood of adopting electronic voting machines

you mean touch-screen?  because we already have optical scan (of paper  
ballots) and i am convinced that is the best method.  nonetheless, it  
is an independent issue regarding which of plurality, IRV, Condorcet,  
Approval, or Range.

> 	• Even if IRV is still an improvement over plurality after tactical  
> voting and these side effects have been taken into account, the  
> adoption of IRV is a loss when you factor in the opportunity cost of  
> adopting IRV instead of Score or Approval Voting (or maybe even  
> Condorcet methods). Reform energy and manpower is extremely limited.  
> If you use limited resources to make 100 dollars when the same  
> resources could have been used to make 1000 dollars, you didn't gain  
> 100 dollars, you lost 900 dollars. That's basic economics that most  
> voting reformers seem blissfully ignorant of.
>
> Now consider IRV's tendency to "backslide" (e.g. in Burlington and  
> Pierce County WA), which presumably leaves voters reluctant to try  
> another alternative voting method, leaving us stuck with Plurality  
> Voting -- and then getting IRV was an even bigger loss.

i've been saying that since March 2009.  the mistake of both  
proponents and detractors of IRV was the insistence that Preferential  
Voting (the ranked-order ballot) is synonymous with the IRV (or STV)  
method of tabulation.  FairVote did the Ranked Ballot a disservice by  
*only* associating it with IRV when they introduced and promoted the  
reform to political jurisdictions.

> So I am not trying to sound like an idiot making some simplistic  
> statement like "IRV is worse than Plurality Voting". It's a lot more  
> complex than that. But the above explanation doesn't make a great  
> soundbite.
>
> -Condorcet: complexity.
>
> i don't see it.  Condorcet is simple and defaults most directly to  
> the "simple majority" rule of two-candidate elections.
>
> Well, it depends which Condorcet method you use of course.

Ranked Pairs is pretty simple.  even for Yahoo's that complain about  
IRV being "too complex".  and meaningful.  and, in the case of a CW or  
a Smith Set of 3, will always elect the same person as would Schulze,  
which seems to be the generally agreed "best method" with Condorcet.   
Schulze is, in my opinion, too complex to explain to and be acceptable  
to the Yahoos.

> But think about the logistics of tabulation with Condorcet.

you need a computer, just like you would for anything else  
(particularly Range).  but Condorcet is precinct summable and all the  
precincts need to do is come up with two subtotals for every pair of  
candidates.  that tabulation is simple and they can report that *both*  
officially up the ladder to the central election authority who is  
deciding the election (and may have to use Ranked Pairs or Schulze or  
something) and to the various candidate campaigns and to the media.   
if there is a CW, it is simple to determine the winner.

> Instead of being able to have precincts sums like Smith=1023,  
> Jones=764, etc., you have to have a table with the pairwise defeats.  
> Not rocket science of course, but certainly more complicated than  
> Plurality/Score/Approval.

only in quantity, but it is not qualitatively different like STV is.   
you *still* have precinct subtotals.  instead of N subtotals, you have  
N*(N-1), which is a lot better than

    N-1
    SUM{ N!/n! }  =  floor( (e-1) N! ) - 1
    n=1

which is the case for IRV (with 4 or more candidates, you may as well  
simply upload every single ballot).

> This may be a relatively minor issue compared to outright  
> performance, but if it causes Condorcet to be political less  
> feasible, or more likely to backslide once implemented, that's a big  
> risk.

well, if you *never* implement it, then what's the point of the risk?   
i mean it's like the risk of a cancer treatment (with promise) killing  
the patient when the cancer certainly will.

>
> While the basic idea of one-on-one matches is simple, the details of  
> tiebreakers are enough to make most voters' eyes glaze over.

then i said:

> by "tiebreakers", do you mean methods to resolve a Condorcet cycle  
> or paradox?  i would agree that Schulze (which i have nothing  
> against, in fact i think it's the fairest way to do it) would make  
> most voters' eyes glaze over.  but i don't think that is the case  
> for Ranked Pairs which is almost as good as Schulze and will elect  
> the same candidate virtually every time, and i would be happy to  
> accept a suboptimal, but simpler, Condorcet (like elect the  
> candidate with the most 1st choices in case of a cycle) just to  
> *get* Condorcet adopted.  small price to pay, because i really am  
> not convinced that cycles will happen very often at all.
>
> Cycles seem to be statistically pretty common actually. Look at  
> table 1 here. It's just a simplistic random elections model, but it  
> seems a reasonable approximation.
> http://scorevoting.net/RandElect.html

it requires a large number of Nader voters to vote for Bush over Gore.

> It actually may be that cycles won't happen, because so many voters  
> will naively exaggerate/polarize the two front-runners that you  
> still have duopoly. But in that case, you approximately have  
> plurality voting. And if the DH3 pathology is real, then you  
> potentially have something much worse.

you can call it a pathology, but if a majority of voters agree that DH  
is a better choice than any one of the 3, then the democratic thing to  
do is elect the DH.  if they elevated the DH in order to bury 1 or 2  
other major candidates, they should expect that their expressed (but  
insincere) preferences will count over their secret (but sincere)  
preferences.  that is not the same strategy problem (compromising)  
that is common in FPTP and has been shown to exist in the 2009  
Burlington IRV.

> There is a lot of speculation here, but it still seems that Score/ 
> Approval are much less risky.

risky of What??  electing the "wrong" candidate?  that's the whole  
issue.

the problem with both Score and Approval is that it places a burden  
immediately upon the voter to decide how much to score each candidate,  
or with Approval, whether or not to approve of candidates that they  
think are not Satan, but neither are their favorite.  the problem is a  
ballot and tabulation method where these "merely approved" of  
candidates are competing against your favorite.  if you don't want to  
harm your favorite, you must consider scoring those "okay" candidates  
with zero (or not approving them).  but then, if your favorite isn't  
gonna win, you would like to help the "okay" candidates beat Satan.   
that is a burden of strategic voting placed upon the thoughtful voter  
the minute they step into the booth.

the simple Ranked Ballot does *not* place that burden on the voter.   
ranking an "okay" candidate #2 or #3 does not harm your #1, at least  
as far as voter expression is concerned.  and your #1 choice does not  
help your #3 beat your #2.  then the problem is getting the tabulation  
method to comply with that voter expression.  *if* there is a CW, then  
Condorcet *does* comply with this voter expression and, i believe,  
*only* Condorcet does.  simply from it's definition.  it is *only*  
when there is no CW that there are conceptual issues where your 2nd  
choice can maybe hurt your 1st.  or where your 1st choice (who is a  
loser) can maybe hurt your 2nd choice against your 3rd.  but that's  
only if there is no CW.

> Even if they somehow aren't quite as good as Condorcet on average,  
> they have better worst-case scenarios, and are just so much simpler  
> and easier to adopt from a cost and logistics standpoint.
>
> Moreover, the need to individually rank numerous candidates is more  
> work than many are ready for,

it's less work than scoring each.  and it's less work than deciding if  
you "approve of" your #2 or #3 choices.  with Condorcet, equal  
rankings are allowed (and, like IRV, every candidate not ranked is  
assumed to be tied for last place).  so with Satan from hell, you  
don't rank him.  then above Satan, you rank Hitler last of all.  you  
rank Jesus as #1, Gandhi and Mother Teresa as #2, and all of the other  
candidates you get to rank equally as #3 because you have no opinion  
of them.  all you are saying is that if Candidate A is ranked higher  
than Candidate B, you would vote for Candidate A if the race were  
between A and B.  you are saying *nothing* more than that and the  
tabulation method should assume nothing more.



then i said:

> there should always be ballot access laws.  there should never be  
> more than 4 or 5 candidates on the ballot (along with Write-In).  if  
> there are, the ballot access laws need to be more strict (more  
> signatures required).  having 20 candidates on the ballot for a  
> single seat is ridiculous.
>
> If you think it's ridiculous, you don't have to vote for any of the  
> ones you don't care about.

right, but there might be candidates that you don't care about that  
are also running against Satan.  you might want to help them beat  
Satan, but you don't want to help them beat your favorite.

> Why should you get to deny me (some other voter) the fullest range  
> of choices?

because of the logistics of a practical ballot in a practical  
election..  maybe we should use the phone book for our ballots?  with  
electronics, we don't even need a ballot with names: we can let  
*everyone* be a write-in.  let's see who wins then?

> And these signature requirements seems to actually prevent un- 
> moneyed candidates from competing.

in the State of Vermont, you need 500 signatures to run for Governor  
or any other statewide election.  100 for State Senator and 50 for  
State Representative.  i know we're a small state, but no one deserves  
to be Governor (or any of those other offices) without drumming up  
*some* initial support, just to show they are viable.

> For instance, here in San Francisco we get zillions of derelicts  
> walking around getting signatures for ballot initiatives, paid  
> something like a dollar per signature. It becomes less about  
> democracy and more about money.

sounds to me, that the problem is the *money*, not the ballot access  
requirement.

> I agree that it can be a bit frustrating and overwhelming when I see  
> more names on a ballot than I could ever have time to research. But  
> with average-based Score or Approval Voting, that wouldn't need to  
> be frustrating, as I could just vote on a random subset of them, or  
> based on name recognition. If a lot of voters did that, the final  
> results would be, in a statistical sense, highly similar to the case  
> where all voters researched and scored all the candidates. That's  
> better to me than having some signature requirement that makes part  
> of the decision for me.
>
> -Approval: divisiveness.
> ...
> -Range: Strategy is too powerful.
>
> i couldn't get the guys at ESF to even acknowledge the obvious  
> strategic considerations a voter would face with Approval or Range.   
> they just say that "it's mathematically proven" to be better than  
> anything else.  Clay Shentrup needs to get on this list and start  
> defending his position rather than expecting me to do the same on  
> his list.
>
> Clay, i'll take you on here on EM, but not on ESF.  it takes too  
> much time and is a far less objective context.
>
> Of course we acknowledge the strategic considerations that Score/ 
> Approval Voters face. We have like 20-30 pages that discuss it. The  
> point is, Score Voting still outperforms all feasible rivals with  
> any conceivable disparity in tactical voting. That is, Score Voting  
> performs better than rivals, even if handicapped with a higher  
> fraction of tactical voters. This is explained in these pages (plus  
> others I'm not even listing here.)
> http://scorevoting.net/StratHonMix.html
> http://scorevoting.net/ShExpRes.html
> http://scorevoting.net/RVstrat1.html
> http://scorevoting.net/RVstrat2.html
> http://scorevoting.net/RVstrat3.html
> http://scorevoting.net/RVstrat4.html
> http://scorevoting.net/RVstrat5.html
> http://scorevoting.net/RVstrat6.html
> http://scorevoting.net/RVstrat7.html
> http://scorevoting.net/Honesty.html
> http://scorevoting.net/HonStrat.html
>
> So, please, do not tell me that we are somehow oblivious to the  
> obvious tactical opportunities inherent in Score Voting. We make no  
> attempt to deny them whatsoever. We simply point out that, even if  
> you take them into consideration, Score Voting is still better than  
> the alternatives.

but you still haven't answered my basic questions of "how much should  
i score candidates that are not my favorite, but are not my least  
favorite"  which includes whether i should "approve" them or not.

you have to get past that basic issue before i bother to go through a  
bunch of propaganda on your website.

also, Clay, *please* post this to the EM list (then it gets routed to  
the right mailbox rather than my general Inbox).  please don't just  
send it to me, personally.

--

r b-j                  rbj at audioimagination.com

"Imagination is more important than knowledge."