[EM] replies to various people re IEVS simulations

[Warren Smith]

>Scott Ritchie:
So, wait, was half the Condorcet electorate strategic voting by doing
order reversal?  Are we making the assumption that strategic voting is
exactly as common in range and Condorcet in these simulations?

That seems a bit strong, exactly because the risks are different and the
information required is greater for Condorcet.

--WDS:
In IEVS, presently, equal rankings are forbidden in rank-order methods.
Hence all strategic voting in them involves some order changing whenever the strategic
vote is not the honest one.

I would not say I am "making the assumption that strategic voting is
exactly as common in range and Condorcet in these simulations."
You are free to make that assumption if you want.  I am simply computing
the data of what the Bayesian Regrets (and probabilities of electing true-Condorcet-Winners)
of different election methods are, at different honesty-strategy mixes in the voter population.
At 50-50 mix, range leads to higher CW probability than Condorcet methods.

You also are free to, e.g, compare Condorcet with a 70-30 mix versus Range with a
50-50 voter mix, or whatever.  Neither I nor IEVS is stopping you.

I have in other work attempted to measure how often actual human voters act strategic.
That is an independent question.  One answer is that human range voters are experimentally
much more honest and much less strategic than human plurality voters.
(Concerning what happens with IRV and Condorcet, I do not really know, but
I speculate the dishonesty fraction in IRV is also high to explain why IRV countries 
all have massive 2-party domination in all IRV seats, whereas countries with delayed runoff,
generally do not exhibit 2-party domination.  Note that even strategic voters are honest in
2-candidate runoff votes.)

>Brian Olson:
It's measuring the wrong thing. Isn't that performance curve about  
the same as without any strategic players? Good methods get good  
answers, even in the face of adversity, ok, BUT

Do the strategic voters make out unfairly well vs the honest voters?

That's what I designed IRNR to solve...

...way back 4-5 years ago. I designed a simulator that could  
measure the social utility of election results, and naturally the  
best result came from the election method which just summed up  
voter's personal-utility-votes and picked the overall best. That's an  
awful lot like ideal range voting. And indeed it's great and  
expressive and better than Condorcet _when everyone is honest_.

--WDS:
I happen to think Bayesian regret is a good metric, in fact the uniquely right metric,
not "the wrong thing".  See
   http://rangevoting.org/BayRegDum.html
for discussion.  However, if you have some other metric you would like to
advance, such as "unfairness" (whatever that is) for strategic versus honest voters
(whatever they are - it is rather hard in practice to tell who is who, they all
say they are honest if asked; and nothing stops voters from providing votes which include both
honesty and strategy inside the same vote) then if you provide a good definition of it I may
include it in a future version of IEVS.  I want to incorporate help and advice from you all.

I will say, though, that personally I really
do not give a damn about how "unfair" an election method is (and indeed am
not even sure what "unfairness" even is); what I care about is how much society benefits
or not, quantitatively, from the election result.  If election method A
causes expected societal benefit +999 and B causes benefit -999, where "benefit" is
a well defined quantitative thing that we all agree is better to make larger -
then I will prefer A even if some whiner comes along and says in his opinion A was 
"more unfair."

As far as how range voting feels awfully close to ideal utility voting (which has
zero Bayesian regret), that is true but it is not a deficiency of range, it is a virtue.
Also, the idea that I "designed" range voting "to the benchmark," falling into
a trap, is nonsense because range voting was invented and used (e.g. in the olympics, and by
honeybees millions of years ago, see  http://rangevoting.org/ApisMellifera.html )
well before the Bayesian Regret concept for voting measurements was invented and
used.  In fact, well before humankind even existed.  

Also, if you are about to now come with the opposite attack that the benchmark was "designed 
to range voting," that idea is also bullshit because the first 
authors of Bayesian-regret-and-voting studies (Merrill, Bordley, Weber) 
had apparently never heard of range voting, or at any rate did not include it in their 
sim studies or writings.  So I hope I will hear no more of this claptrap.
Sometimes a test actually reveals something and nobody faked it.  Astonishing
news, but it can happen.

Regarding IRNR, if Brian Olson sends me code for it, then I will be very happy to add it to
IEVS also.  If anybody else has a voting method they want me to add to IEVS, please send
me code.  Or good pseudocode.

Finally as for "Isn't that performance curve about
the same as without any strategic players?" I don't know what that meant since
I do not know what a "performance curve" is.
However, I will say that some voting methods, according to IEVS, perform much
better or much worse relative to the others if we alter the strat/honest mix.
For example, Borda is among the best with honest voters but drops like a rock with
strategic voters (to about midpack).   Approval+top2Runoff is the best with strategic voters,
but with honest voters it drops like a rock down to about 19th place.
Range happens to be, pretty much uniquely, good across the board no matter
what the voter mix.

I will soon emit another post explaining current IEVS status which hopefully will make
clearer some of its (1) findings, (2) strengths, and (3) deficiencies.
IEVS has many deficiencies but it is (by a lot, far as I know) 
the best voting simulator in the world.
So it would be my recommendation to help IEVS evolve further to overcome its deficiencies.

Warren D Smith
http://rangevoting.org
warren.wds AT gmail.com