[EM] Graphics of simulations

[Forest Simmons]

On Mon, Feb 27, 2023, 4:35 PM Richard, the VoteFair guy <
electionmethods at votefair.org> wrote:

> On 2/26/2023 9:47 PM, Forest Simmons wrote:
>  > Here's my question: do simulations carry any weight
>  > with the public? Or do they just care about choice
>  > of buzz words and phrases like democracy; majority
>  > rule, etc?
>
> I have not seen hardly anyone in the "public" realm who initially seems
> to care about simulations.
>
> Mostly they care about "who will win?" "which party does it favor?"
> "does it empower minorities?" and most of all: "Can I understand it? And
> does it make sense?"
>

These two questions should be important to everyone ... especially method
designers.

Another important question is efficiency and transparency of tally... a big
weakness of IRV.

Also ... is sincere strategy optimal? Or do voters have to reverse their
ballot preferences to protect their sincere preferences.

>
> For the latter, the FairVote organization has popularized the notion of
> eliminating one candidate at a time.  It's easier to comprehend.
> Especially compared to something like the Smith set.  Even the Condorcet
> winner concept is difficult for many, many voters to understand.
>
> Aside: I've had success educating "the public" about the idea that a
> "pairwise losing candidate" -- which is a simpler variation of
> "Condorcet loser" -- deserves to be eliminated.  A soccer analogy helps:
> if a soccer team loses against every other team then it shouldn't be
> possible for that team to win.  (This is one of the two refinements over
> IRV that the RCIPE method, mentioned below, offers.  The other
> refinement is to count ballots that the FairVote folks wants to discard
> as an "overvote.")
>

Great work!

What did you think about the Martin Harper solution to the "over vote"
problem?


> Yet simulations can be useful _IF_ the results are presented as a
> _GRAPHIC_ that compares "your" methods with familiar methods.
>
> I created a graphic (at the following link) that shows failure rates


What constitutes "failure"? If we have notions of success and failure that
appeal to voters we can design our methods around them. Which criteria of
success and failure measures by your graphics appeal the most to them?

for
> RCIPE (ranked choice including pairwise elimination) and IPE (instant
> pairwise elimination)


How is IPE received? That is the focus of my recent efforts ... nominal
elimination criteria backed up with pairwise confirmation when possible ...
else pairwise over-ridden ... in a way that makes burial, chicken
defection, and other insincere manipulations likely to backfire.

Does the public care about that?

I ask because that is the main focus of our simulations.


> plotted along with IRV, plurality, Borda, and
> Kemeny.  That graphic has been helpful to some people who already
> understand IRV and have heard about Condorcet methods.
>
>    http://votefair.org/clone_iia_success_rates.png
>
> In this sense such a graphic is like a Yee diagram.  Most people
> intuitively recognize that the Yee diagram for IRV reveals that method
> has serious flaws.  (However, trying to explain Yee diagrams is not
> fruitful.)
>

Too bad ... it's pretty simple as far as graphics go ... the color of each
pixel is the color that identifies the candidate who would win if the
voters were distributed symmetrically about that pixel.

It's a huge mistake to introduce two dimensional Yee diagrams without first
getting students comfortable with one dimensional Yee diagrams.

If you do a good job in the one dimensional case, the two dimensional case
is pretty much optional ... Galileo didn't need the Hubble telescope to get
the basic picture ... nor did Van Leeuwenhoek need an electron scanning
microscope.


> The fans of STAR voting have gotten lots of mileage from their graphs of
> VSE (voter satisfaction efficiency).  Alas most people don't realize
> that the "efficiency" is meaningful among a group of friends or
> cooperative people, but is not meaningful in governmental elections
> where strength of opinion violates the principle of "one person one vote."
>

But STAR is a runoff like IRV in which the winner is determined by which
candidate gets a majority preference in the final pairwise contest. Each
participating voter has one vote in that contest ... all of the previous
stages are clumsy jockeying to see who the two finalists are for the actual
one-person-one-vote contest.

But, you say, "Every stage of IRV serves to shuttle the voter's 'vote' from
the first stage to the final stage."

But those stages form a very lossy channel for the "vote" signal.

Those stages are plagued by vote splitting, center squeeze, etc.... Without
the pairwise check of RCIPE they have very little democratic validity ...
just superficial correlation with suitability for continued participation
versus elimination.

[Martin Harper invented his method to disentangle the ultimate over-vote ..
approval itself, to find the logical destination of the "one vote" of the
"one person" who voted a given approval ballot. His solution was to add it
to the total of the ballot approved candidate with the most support from
the other voters. Your one vote goes to the candidate you approve with the
best chance of winning. The generalized Martin Harper rule accomplishes the
same disentanglement for score ballots ... to give them "one ballot, one
vote" compliance ... a way of deducing who "got the vote" of the person who
submitted the ballot.]

As far as unequal vote strength goes ...

Just as score gives more strength to high ratings .. so IRV gives more
strength to high rankings ... which (in both cases) is a big incentive for
insincerely ranking compromise over favorite ... the important urgency of
first place rankings and ratings (and their role in subverting sincere
voting) cannot be ignored.

That urgency wouldn't be there if the IRV big lie were true ..."go ahead
and rank your favorite first ... if it doesn't win, your vote will transfer
to your second choice."

RCIPE is an important corrective ... but other, more streamlined IPE
variants are much more efficient, transparent, and practical... if only the
public could understand that reality.

In score voting that compromise urgency makes score strategically
equivalent to Approval ... strategically it would be just as good to do
STAR with RCV ballots and approval cutoffs ... and have the runoff between
the top two approval candidates.

In that case there would be zero incentive for insincere rankings.

In fact, any runoff method that picks the two finalists with a provisional
set of ballots (approval, score, grade, judgment, RCV, whatever) and
reserves a fresh set of RCV ballots for choosing between the two finalists
... gives zero incentive for insincere rankings on the final ballots.

Any voters that want to abstain from that messy, provisional, strategical
ballot should be able to do so, imho.

What do you think about the VPB Vote for a Published Ballot option?

It might get more participation from voters put off by the inconvenience of
copying those recommendations onto a ballot in the voting booth.


> In summary, graphics that show many thousands of simulations for
> multiple methods are useful among informed voters.
>
> Such graphics are also useful for us, the experts.  It reveals the
> extent to which a method is better than other methods.
>
> In other words, just looking at specific cases to identify failure
> possibilities is of limited use.  In contrast, quantitative graphics
> that allow failure rates to be compared with familiar methods are quite
> useful.
>
> Not as "proofs" but as supporting evidence for claims about being "better."
>
> And such graphics answer the question "Is this 'better' method worth the
> extra effort it takes to understand it and calculate it?"  If there is
> only a small gain, most people will say "no it's not worth the extra
> complication."
>
> Forest, I've enjoyed your speculations about better methods.  It would
> be interesting to see graphics that show specific failure rates -- such
> as IIA, clone independence, burial resistance, chicken resistance, etc.


That's exactly what we need. Kevin and Kristofer have created some top
notch graphics along those lines to satisfy our curiosities ... but
somebody needs to digest them and adapt them for public consumption.

Such graphic comparisons will reveal whether your improvements are big
> improvements or tiny improvements, and which methods excel at which
> characteristics.
>
> Then it will become easier to find a balance between mathematically
> ideal and "good enough" for use in real elections.
>
> Richard Fobes
> The VoteFair guy
> ----
> Election-Methods mailing list - see https://electorama.com/em for list
> info
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20230227/9d7ea057/attachment-0001.htm>