Simplified VSE graph

[Jameson Quinn]

Here's a graph I just made of VSE for just 5 key methods: plurality, IRV, approval, SRV, and 3-2-1 voting (0-10). Do what you want with it.

[Nevin Brackett-Rozinsky]

Am I reading correctly that in your “100% strategic” model, plurality achieves substantially better results than IRV?

[Clay Shentrup]

Approval does great with strategic voting, a tiny bit better with 50/50 strat/hon voting, and WAY worse with 100% honesty. Something seems drastically off in your model.

[parker friedland]

Can you include Minimum Median Max:
https://groups.google.com/d/topic/electionscience/STeXZE2Lr5g/discussion

[parker friedland]

Or Majority Judgement

[Jameson Quinn]

Clay is surprised that

 

Approval does great with strategic voting, a tiny bit better with 50/50 strat/hon voting, and WAY worse with 100% honesty. Something seems drastically off in your model.

 

That's not wrong. Approval with strategic voting is essentially approval with runoffs; people are honest in the polling, then set their threshold based on the (imperfect media version of) the frontrunners.

Warren's BR "strategies" were all based on completely arbitrary frontrunners. It was a start, but it's wrong. Yes I understand that we've all been thinking for the last 15 years that strategy could only make things worse, but it turns out that's not true.

To explain what my model does with "media": I take the totals or rankings from the first round, find the standard deviation, then add gaussian noise to each number with the same standard deviation. So the 2 polling frontrunners are usually correct, and if not usually in the top 3 of the 6 candidates I use, but not always.

2017-02-26 2:25 GMT-05:00 Clay Shentrup <cl...@electology.org>:

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Jameson Quinn]

Yes. But that's because the strategy I'm using in IRV is optimized for "I'm the only one strategizing". If there are multiple candidates who can beat the honest IRV winner, the strategy will elect the least popular of them. In other words, I think that "100% strategic" is a thing that could actually happen in plurality, but not a likely thing in IRV, because normal humans would never think of the strategy I'm using for IRV.

--

[Jameson Quinn]

I included 2 median methods in my full graph: http://electology.github.io/vse-sim/VSE/

--

[Warren D Smith]

I don't think it was "wrong" of me to make the 2 frontrunners be
completely arbitrary. Although it might have been "an exaggeration of
the problem."

1. Consider the latest US presidential election. Trump became
a frontrunner with plurality voting in the GOP primary. The
reason that happened was not because
of some honest-voting polls plus added noise. The reason it happened
was because Trump was (a) famous and (b) different than all his
rivals who by comparison were all "nearly the same." Trump intentionally
loudly and eternally emphasized a & b.
The facts that Trump was the most unqualified,
oldest, and most-interest-conflicted, person
ever to become US president and ran with essentially no agenda
while telling more lies than any major candidate in the history of fact-checking
agencies, all were irrelevant. In fact most of that helped create more "b."
Therefore, AS A CONSEQUENCE of strategic voting behavior from then
onwards, he sailed to the nomination easily.

OK? Trump became a frontrunner for reasons which as far as any
simulation should be concerned, were
COMPLETELY ARBITRARY and had little or no relation whatever to
honest polling plus noise.

2. It often happens the anointed "frontrunner" is thus-anointed purely
because of press reports that he has accumulated an enormous amount
of money. E.g. Jeb Bush in 2016 whose campaign went nowhere
in spite of this anointment.

Is this due to "honest polling plus noise"? Well, it seems quite
arbitrary or maybe even worse-than-arbitrary in the sense that it
favors more-corrupt candidates.
I suppose you could consider the rich guys giving him the early money to be
thus "polled" but I would contend they are so far far off being
a representative sample that they have to be
viewed by a simulator as totally arbitrary.

3. If often happens that somebody wins Iowa and/or New Hampshire, then
is branded a "frontrunner" then as a consequence steams to the nomination
while meanwhile somebody who seems comparable or more qualified
etc is branded a "nonfrontrunner" due to poor Iowa
performance and thereafter dies the death.

Well, is that the result of "honest polling plus noise"? Well, certainly
arguably yes, if you regard the Iowa caucus as
some sort of honest polling plus noise. But there is a hell of
a lot of noise.

So rather than branding me as wrong purely based on your intuition, you
should actually consider the evidence from the real world.

But I think I like JQ's "honest polling plus noise" idea in principle for
determining "frontrunners," and note it becomes the same as my method
in the limit of high noise. I just want to say that
(a) my limit is not at all bogus.
(b) it is very simple to program and get right.

But the JQ approach is NOT simple to program and get right, in fact not
even easy to think about or define right. The more you think about it,
the deeper waters you find yourself swimming in.
To explain what I mean by that:
If you have an arbitrary (perhaps very complex)
voting method which has to be treated as a "black box"
inputting votes and outputting winners...
well, suppose we feed it some honest votes and it says "A wins."
Now what? Who should be the "frontrunner?"
Well, there may be no such thing -- for a general black box -- as a "2nd placer"
B so we have no general-purpose way to select "B" with greater chance than C.
And ok, what if we say "well, we'll feed in NOISE-DISTORTED
honest votes and then the winner will be somewhat random"
and we'll declare him frontrunner.
Well, then the problem is different voting methods
have different inputs, e.g. approval, ratings, or rankings,
also other things too are possible. In general the input is just a bitstring.
So if you try to "add random noise" to distort the honest votes,
you are doing something different for each
voting method, which might be unfair and comparing
apples to oranges.

Eh? So then you get into murky waters.

My way, it is at least transparently obvious it is totally apples-to-apples
and no definitional confusion and murkiness.



--
Warren D. Smith
http://RangeVoting.org <-- add your endorsement (by clicking
"endorse" as 1st step)

[Jameson Quinn]

2017-02-26 13:13 GMT-05:00 Warren D Smith <warre...@gmail.com>:

I don't think it was "wrong" of me to make the 2 frontrunners be
completely arbitrary.  Although it might have been "an exaggeration of
the problem."

1. Consider the latest US presidential election.  Trump became
a frontrunner with plurality voting in the GOP primary.  The
reason that happened was not because
of some honest-voting polls plus added noise.  The reason it happened
was because Trump was (a) famous and (b) different than all his
rivals who by comparison were all "nearly the same."  Trump intentionally
loudly and eternally emphasized a & b.
The facts that Trump was the most unqualified,
oldest, and most-interest-conflicted, person
ever to become US president and ran with essentially no agenda
while telling more lies than any major candidate in the history of fact-checking
agencies, all were irrelevant.  In fact most of that helped create more "b."
Therefore, AS A CONSEQUENCE of strategic voting behavior from then
onwards, he sailed to the nomination easily.

OK?  Trump became a frontrunner for reasons which as far as any
simulation should be concerned, were
COMPLETELY ARBITRARY and had little or no relation whatever to
honest polling plus noise.

I don't think that Trump being a frontrunner was completely arbitrary, nor do I think that it was entirely honest. In polls, I think most people honestly gave their favorite among the set of candidates they took seriously. Trump's fame and apparent riches gave him an advantage in being taken seriously, but once he made it onto people's mental list, they either liked him better than the alternatives or they didn't. So I think "honest plus noise" is a reasonable approximation.    

This is a good point; I entirely agree.

 

I just want to say that
(a) my limit is not at all bogus.
(b) it is very simple to program and get right.

But the JQ approach is NOT simple to program and get right, in fact not
even easy to think about or define right.  The more you think about it,
the deeper waters you find yourself swimming in.
To explain what I mean by that:
   If you have an arbitrary (perhaps very complex)
voting method which has to be treated as a "black box"
inputting votes and outputting winners...
well, suppose we feed it some honest votes and it says "A wins." 

Now what?  Who should be the "frontrunner?"
Well, there may be no such thing -- for a general black box -- as a "2nd placer"

This may be a problem in theory, but in practice all the systems I tested output either scores or a rank order. In some cases, there was more than one way of generating a rank order that would be compatible with the winner; in such cases, I picked the way that seemed to me to be most "natural", but I could imagine some debate. But since I'm adding noise anyway, and since in most cases all reasonable ways of generating a rank order will agree on most candidates, I don't think that's a big issue. 

 

B so we have no general-purpose way to select "B" with greater chance than C.
And ok, what if we say "well, we'll feed in NOISE-DISTORTED
honest votes and then the winner will be somewhat random"
and we'll declare him frontrunner.
Well, then the problem is different voting methods
have different inputs, e.g. approval, ratings, or rankings,
also other things too are possible.  In general the input is just a bitstring.
So if you try to "add random noise" to distort the honest votes,
you are doing something different for each
voting method, which might be unfair and comparing
apples to oranges.

Eh? So then you get into murky waters.

My way, it is at least transparently obvious it is totally apples-to-apples
and no definitional confusion and murkiness.



--
Warren D. Smith
http://RangeVoting.org  <-- add your endorsement (by clicking
"endorse" as 1st step)

[parker friedland]

Can you create a version of that graph that resumes the median satisfaction produced by each of the voting methods instead of the average?

If it is possible to get voter’s honest satisfactions of candidates by sticking dopamine detectors into their heads, choosing a candidate with a great median satisfaction is generally better then choosing one with a great average satisfaction. This is because when you chose a candidate with the greatest median support, you tend to choose a candidate that pleases a greater number of people then by choosing the candidate with the greatest average support. By doing this, you also tend to pick more centrist candidates because centrist candidates tend please the greatest number of people while candidates closer to the edge of the political spectrum tend to get propped up more by concentrated support for their own side of the political spectrum then they would in median based voting systems.

I understand that which type of measurement is the best, whether it's median or mean, is dependent on one's philosophy. However it doesn't hurt to include a median based VSE graph.

[Clay Shentrup]

On Sunday, February 26, 2017 at 6:51:39 AM UTC-8, Jameson Quinn wrote:

That's not wrong. Approval with strategic voting is essentially approval with runoffs; people are honest in the polling, then set their threshold based on the (imperfect media version of) the frontrunners.

What's strange is that your results are apparently a hill, where strategic/honest mixture is better than all strategy or all honesty. This seems very very strange to me. I'd like to see the hill in graph form. 

[Warren D Smith]

Also, Jeb Bush was anointed a frontrunner due
to his fame and money, which in turn were mostly
due to him being named "Bush."

Which was a random event.

If he'd been the same guy, same career,
but been named "Jomamba Papagopoulis,"
then he would have been a lot less famous,
had way less big money in his chest, and
the press probably would not have anointed him
as "frontrunner," at least not nearly so early.

And no, I do not think Trump getting frontrunner status
was "partly honest." He got that way well before his web page,
for example, contained any "issues" section at all.
He had no issues at all, and he is the frontrunner.
It was simply due to his fame, also called "name recognition,"
which in turn was due to his TV show and general
lifelong playboyitude. Similarly for Arnold Schwarzennegger
becoming "frontrunner" and therefore easily winning. That
was a random event ascribable to his fame+wealth ascribable to
his having a lot of muscles. It wasn't due to his political
qualifications and stances on issues, of which he had none.
Perhaps he later did, but only well after achieving frontrunner status.

> Trump's fame and apparent riches gave him an advantage in being
>taken seriously, but once he made it onto people's mental list, they
>either liked him better than the alternatives or they didn't.

--that is confused. The fact is, "they didn't"; pairwise polls indicate Trump
would have been defeated pairwise by, e.g. Kasich.
The problem was, Trump's opposition was split so there was no pairwise
race.

The thing is, once somebody gets to be frontrunner, they stay that way because
it is unstrategic for anybody to ever tell a pollster any choice
besides the top 2. So as long as Schw or Trump could keep their rivals split,
they had an easy pre-ordained win.

Also, another thing is, Iowa caucuses are actually a DIFFERENT
voting system than the rest of the primaries etc. So if you become frontrunner
because you do well in Iowa, is that well modeled by a random sample
poll with honest voting? NO, because it was a different voting
system. Also a pretty different sample. So becoming frontrunner
by winning Iowa, is from the point of view of some other voting
system, a pretty random event.

This all had occurred to me when I wrote the sim back in 1999-2000
and this was the sort of reasoning I used to justify the way I did it.
I had considered making the frontrunner be based on a subsample
poll but then there would be more knobs to fiddle with
and more murkiness and apples-to-oranges issues, and
the "right" setting of the knob would be (as you put it)
"totally arbitrary" and hard to defend anyhow, so why not
just go with the knob setting "infinity" which is simple and easy.
And I still think that has plenty of validity to it. And you will notice
I have a preference for keeping things simple and clearly-defined,
given a choice.

But anyhow I think I do agree that in principle the optimal knob setting
(whatever it is) must be better than "infinity", but is the advantage
worth the can
of worms that must be opened to enjoy it?

[Warren D Smith]

And another argument of the same ilk pertinent to a lot
of elections at well below the presidential level (e.g. statehouse)
is: the Dem and Repub are near-automatically granted
"frontrunner" status on day 1. This is a pretty random
event attributable in many cases to a single party boss,
or to somebody having the right friend. If the same exact people
had run but with the party-affiliations (e.g. "Green" and "Republican")
switched, then somebody else would
get that frontrunner status.

Entirely a random event, having virtually nothing to do with honest polling
of a subsample with noise.

[parker friedland]

The comparison between ranked pairs and Schulze beatpaths is very interesting. I would of thought that Schulze would do better because it produces winners that don't tend to have as huge of a defeat as with ranked pairs when there isn't a Condorcet winner. This is perplexing.

[parker friedland]

I would have to assume that the differences between score 0 to 1000, score 0 to 10, score 0 to 2, and ideal approval when 100% of the voters are strategic is just a result of the randomness of types of voters in each test. All of these values should be the same because the most strategic way to vote in range is with only max stars or min stars for each candidate. If this is the case and the results are not that accurate yet because not enough simulations have been ran, why are you displaying the different voting system's VSE scores with 3 sig-figs?

[Jameson Quinn]

2017-02-26 23:20 GMT-05:00 parker friedland <parkerf...@gmail.com>:

I would have to assume that the differences between score 0 to 1000, score 0 to 10, score 0 to 2, and ideal approval when 100% of the voters are strategic is just a result of the randomness of types of voters in each test. All of these values should be the same because the most strategic way to vote in range is with only max stars or min stars for each candidate.

But the information based on which you're strategizing comes from honest results plus noise, and honest results are different in each system.

 

If this is the case and the results are not that accurate yet because not enough simulations have been ran, why are you displaying the different voting system's VSE scores with 3 sig-figs?

--