[EM] Yes/?/No

[Kristofer Munsterhjelm]

On 07/11/2020 03.52, Forest Simmons wrote:
> 
> 
> On Friday, November 6, 2020, Kristofer Munsterhjelm
> <km_elmet at t-online.de <mailto:km_elmet at t-online.de>> wrote:
> 
>     On 06/11/2020 06.11, Forest Simmons wrote:
>     > In summary, to the common person, the biggest selling point of IRV/RCV
>     > is that it achieves a 50+ majority without a physical runoff. And its
>     > biggest defect is that it requires the voters to rank the candidates.
> 
>     I'd say the value of that property is rather illusory. For one, voters
>     can change their minds between rounds of an actual runoff, which allows
>     for some strategy beyond what IRV provides. In addition, actual runoffs
>     can also fail to center squeeze. So sacrificing equal-rank and
>     truncation on the altar of "majority winners every time!" does not
>     actually buy the voters what they think it does.
> 
>     > And here in the States we compromise the biggest selling point in
>     order
>     > to ameliorate the biggest defect. So we have to fall back on the
>     second
>     > biggest selling point which is resolution of the Duverger problem that
>     > entrenches the two party dynamic. That's a good selling point for 3rd
>     > party supporters but not easy to get typical voters excited about.
>     >
>     > The good news is that Dodgson gives us a way to break the two party
>     > stranglehold without the use of rankings, ratings, etc.
> 
>     Does it, though? Consider Australia. The how-to-vote cards are, for any
>     voter that follows such a card, in essence a precommitted ranking proxy
>     method on top of IRV. And despite having STV for one of its houses,
>     Australia is still two-party (if you count Lib-Nats as one).
> 
>     It would seem that it's not enough to have a delegation mechanism on top
>     of a voting method; the voting method itself has to not be *too*
>     Duvergerian.
> 
> 
> I did not mean to imply that any old base method (least of all
> IRV/RCV/STV) would do, rather only that any good method could be made
> voter friendly by use of proxies, whether or not it required complete
> rankings or majorities of one kind or another.  

Ah, I misunderstood. I still suspect, though, that what seems to
be inevitable for ranked systems (e.g. the complexity of ranking
absolutely every candidate) may just be a consequence of certain
implementations of certain ranked methods (IRV).

It's important to distinguish between how hard ranked voting methods are
in general as opposed to how hard a particular ranked voting method is -
just like it's unfair to condemn e.g. Ranked Pairs because IRV is a
lousy method, it's also unfair to have the difficulty of ranking every
candidate in IRV with absolute majority requirements count against
Ranked Pairs as a method.

But of course, there's the problem: since AV is the most notable use of
ranked voting (which is how FairVote can get away by calling IRV "ranked
choice voting"), it's hard to disentangle the effects. There's little
data on the use of non-IRV methods.

I have checked with some local NZ election data, since in NZ local
elections that use STV, truncation is allowed (unlike in Australia). In
addition, since some regions use FPTP and others STV, it should give a
pretty reasonable picture of what happens when you use STV or FPTP, all
other things equal.

In 2016 and 2019, in both the mayoral and district council elections (AV
and STV respectively), the turnout for all STV districts as a whole
(i.e. the chance that a random eligible voter in an STV district
actually casts a ballot) is higher than the similar turnout under FPTP.
Spoilage rates are somewhat higher in STV district council elections
than FPTP ones, so if you take spoiled ballots into account, the rates
are about equal. (See document.)

The results would suggest that ranking is not that much more difficult
than voting for one; or, at least, that it's not enough of a burden that
the voters fail to show up.

The source data can be found at
https://www.dia.govt.nz/Services-Local-Elections-Local-Authority-Election-Statistics-2019.

>> In addition, you likely have to spend some kind of political
>> capital to get a complex method through, so the return on the additional
>> complexity may not be worth it.
> 
> There is no complexity to VPR, which is what I suggest for base methods
> like River, CSSD, ASM, RP, et.

The complexity would mostly be in the infrastructure: allowing the
candidates to submit their lists, making sure those are printed,
registering the proxies and substituting the votes in the algorithm, etc.

I guess what I'm saying is that *if* ranking isn't too onerous, then
there shouldn't be much of a reason to use a proxy ordering. The NZ
results suggest that it's not, at least not in that setting. Now,
particular methods may make the burden of voting properly much higher:
e.g. IRV/AV with what you called the "Majority fetish"; or Approval,
where it's hard to figure out which honest vote to submit. But much of
the inherent complexity in River, Schulze, etc. is there to satisfy
criteria that make quick ranking work.

There's another perspective which may be influencing my position: the
parliamentary election system in Norway is (mostly) closed party list.
In local elections, it's possible to give an additional vote to a
particular member of a party's list; but very few people do so, and so
the election results per party per municipality or region generally
follows the list order, which is generally set by the party leadership.
Part of me is concerned that by providing proxy as an easy option, the
system may get locked into a closed list-like dynamic, where the parties
decide and too few voters bother to vote "below the line", as it were.

In any case, I think that the advanced methods are flexible enough to
support low-burden votes, and good enough to do reasonably well given
them[1]. Of course, if the voters of a particular small town *feel* that
ranking is too complex, they may decide to use VPR anyway, but I think
the absent such a demand, there's no need to augment an advanced methods
with such a mechanism. If the closed list experience is generalizable,
there may be a risk to supplying a vote proxying mechanism.

(I'm not sure if any method with grades or approvals would support
low-burden votes, but that might just be because I'm in the minority
(apparently) that finds ranking easier than rating. This because when
ranking there's no need to calibrate the scale.)

[1] Condorcet in particular seems to resist noise quite well; I think
Brian Olson did some simulations on this.

>>     To avoid double counting, in a straightforward Approval Asset, the
>>     tradeable assets should be votes, not approvals. Proxies could only give
>>     a ballot to a candidate who is listed as approved on the ballot in
>>     question, and the candidate with the most ballots after negotiation
>>     wins.
>>     But it's not clear how to initially distribute the votes to the proxies,
>>     unless the voter specifies a favorite. If you split each ballot evenly
>>     between the approved candidates, you get something more like cumulative
>>     vote plus negotiation, and that doesn't sound like a good solution.
> 
> 
> Combining Asset with Approval in the way you suggest below makes the
> approval ballots too stiff ...  not enough flexibility to fully exploit
> the strategic nature of approval voting during the bargaining.

I was thinking of a Republican-Democratic-Green scenario. The idea would
be that e.g. a Green voter can freely approve Nader without (much) risk.
Because the voter's ballot approves of both Nader and Gore, the Green
party can then use the ballot as a chip when bargaining with the
Democratic party. However, a hard-core Nader voter could still choose
not to vote Gore, in which case that ballot can't be transferred to
count for Gore. So the "stiff" approach seems to work.

In a Burlington setting with a number of equally large candidates, it
still comes down to who blinks (concedes) first, but I imagine that's
common to all Approval proxy methods.

It would be possible, under simplifying assumptions, to argue that the
median voter candidate is in a position closest to a compromise that
would please everybody, and so would win such negotiations -- in which
case, Montroll would probably have won. But then again, such assumptions
may not necessarily hold in the real world.

> 
> 
>     And the simplest variant of EMV is probably straight up Approval, but
>     with continuous feedback. The voters approve of a number of candidates
>     and the number of approvals for each is publicly shown. The candidates
>     then adjust their positions according to the approval distribution until
>     some deadline, after which the candidate with the most approvals wins.
>     But it's only simple in theory. The logistics is a completely different
>     matter.
> 
> 
> One more remark about VFA: here's the best way imho to give the punch
> line: once all of the question marks have been resolved into yeas and
> nays, the candidate with the greatest ratio of yeas to nays is declared
> winner. 

Isn't Approval the same whether you use the ratio of approvals to
non-approvals or just the count of approvals? Since the logarithm is
monotone increasing on positive reals, x/(v-x) > y/(v-y) should hold
when x-(v-x) > y-(v-y), which again holds when x > y. So it shouldn't
matter how you phrase it. Or am I missing something?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NZ comparative turnout.ods
Type: application/vnd.oasis.opendocument.spreadsheet
Size: 52518 bytes
Desc: not available
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20201125/ede3176b/attachment-0001.ods>