bad news for voting reform; what voters want

[Warren D Smith]

PolicyInteractive, a nonprofit, did some online surveys / mock elections
during the USA 2016 election (pre & post) using alternative voting
systems.
The most important person involved in that effort was Robin Quirke.
Supposedly their sample was attempting to be representative, but
it wasn't in the sense it was an online sample, and in various
ways this sample differed politically from true random samples.

After filling out mock ballots for all the systems (pre-election poll),
642 were asked: "Which voting systems did you prefer (0 is worst; 5 is best)?"
i.e. they were to rate each system on 0-to-5 scale.
SYSTEM MEAN_SCORE 95%CONFIDENCE_INTERVAL
Plurality 3.89 [3.77, 4.01]
Instant Runoff 2.46 [2.33, 2.59]
Score Voting 2.40 [2.28, 2.52]
Approval 2.02 [1.89, 2.13]

Ouch. This result is not good news for voting method reform, and it
conflicts with some previous surveys, e.g. in France, mentioned here
http://rangevoting.org/WhatVotersWant.html
It agrees with some previous surveys mentioned there that score>approval
and plurality>IRV, but disagrees with previous claims score>plurality
and approval>plurality where ">" means "majority preferred."

Re score vs plurality:
I think the PolicyInteractive survey was done better than the French survey,
except for the fact the former was online while the latter was an exit poll
of genuine voters and was answered by
a high percentage of all genuine voters at their locations.

PolicyInteractive also re-asked a similar system-rating question AFTER
the official election, of the same pollees, and the results were similar, i.e.
score and IRV had overlapping confidence intervals but
both were confidently above approval. I am trying to acquire the pairwise
matrices.

---------

Re who would win the US presidency, it looks like their surveys
all said "Hillary Clinton" #1 and Donald Trump #2
with all voting systems, but her most-clear victory
seemed to be with range and approval voting; whereas with plurality and IRV her
victory seemed within their margin of error. E.g. in their 3-7 Nov survey,
631 participants all of whom scored all candidates,
score voting results shown as mean+-stderror.
And HONSCORE was from the question
"Regardless of their chance of being elected, how much do you honestly
want the following to be elected? Jill Stein."
The final line of the table gives the number of stderrors needed to
bring Clinton below Trump (assuming perfect anticorrelations).

CANDDT SCORE(0-5) APPROVAL PLURALITY HONSCORE
H.Clinton 2.41+-0.0878 0.506+-0.0140 0.489+-0.0142 2.225
D.Trump 2.19+-0.0880 0.475+-0.0144 0.467+-0.0145 2.048
G.Johnson 1.78+-0.0558 0.167+-0.0135 0.034+-0.0071 1.331
J.Stein 1.65+-0.0547 0.136+-0.0127 0.010+-0.0040 1.165
Clint>Trump? 1.25sigma 1.09sigma 0.77sigma

With IRV, Johnson & Stein had by far the lowest top-ranking
counts and hence no question they had to be eliminated, leaving Trump
vs Clinton,
which was won by Clinton in final IRV round by 324-307, where note
324 is really 324+-12.6 putting in +-stderror, which means the Clinton
"victory" with IRV was not confident, because the +-stderror was enough noise
to bring Clinton below 50%. I.e. Clinton's IRV victory over Trump
is an 0.67sigma conclusion.

So in all their voting systems Clinton won, but in none of them
was this victory statistically significant; the only truly
significant conclusions were that Clinton & Trump were the top 2 in
all 4 systems.

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

[Warren D Smith]

Quirke sent me one of her pairwise matrices
(this one is not from the same survey, though)

* Approval IRV Plurality Score Undecided
Approval * 98 53 113 107
IRV 47 * 43 75 107
Plurality 92 102 * 110 117
Score 32 70 35 * 103
Undecided 38 38 28 42 *

note Plurality was preferred versus every rival system pairwise
and there was a transitive preference ordering
Plurality > Approval > IRV >* Score
in this survey, where the >* denotes a preference which
was not statistically significant, at least not in this survey.

Maybe more pairwise matrices will come later.

[Rob Wilson]

It is disheartening that so many people picked plurality. People are just fundamentally really bad at math or they really don't like change.

I'm not surprised about Clinton winning though. I hate her, but if I were the king maker and was the only one to get to decide who the President was going to be of the four, I'd pick her. None of the four candidates were any good. Part of the reason for that is because anyone attempting to run as a third party under plurality has got to be crazy. By running, you are just hurting the candidate you agree with most.

[Toby Pereira]

It doesn't surprise me at all that people preferred plurality over other methods. People are used to it, and they haven't been educated in alternative systems and haven't necessarily considered them in any detail. I don't think it means they are necessarily bad at maths.

[Brian Olson]

I was reading up on IRV failure cases around the country, where it was tried and repealed, and the news articles cited a couple cases of the sentiment of "the candidate who was who was in the lead in the first round didn't win, so we're confused and angry". Probably mostly by supporters of that candidate who felt they were robbed of the win. There needs to be messaging about how this is normal, this is actually kinda the point. I found a group of people working towards a ballot initiative in Massachusetts. My first priority is to try to shift their direction away from the fairvote lies and towards implementing Condorcet/Virtual Round Robin style Ranked Choice Voting. But a very close second seems to be the need for public education about how all this stuff is supposed to work.

--
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.

[Mark Frohnmayer]

I love that they used score voting to determine which voting system was liked most.

--
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 electionscien...@googlegroups.com.

[Warren D Smith]

--I will attempt to obtain a pairwise matrix --
which is deducible from their score voting data,
they just (so far) have not actually performed that deduction...
It is conceivable (though unlikely?) that the PolicyInteractive and
French votign-system-preference
data actually are not in contradiction and this
will be revealed when the pairwise matrix is found.

On 1/1/17, Mark Frohnmayer <mark.fr...@gmail.com> wrote:
> I love that they used score voting to determine which voting system was
> liked most.

--given that there were 4 alternatives, some other systems would
not have worked well...

[Robin Quirke]

Warren, regarding your statement, "Supposedly their sample was attempting to be representative, but  it wasn't in the sense it was an online sample, and in various  ways this sample differed politically from true random samples": This claim was never made by me, in fact in the results paper for this study in the methods section, it reads "the results should not be taken as indicative of population-wide voter representativeness or statistically representative support for alternative voting systems." I am of an average mathematical mind, but I know better than that.

[Warren D Smith]

On 1/3/17, Robin Quirke <robin...@gmail.com> wrote:
> Warren, regarding your statement, "Supposedly their sample was attempting
> to be representative, but it wasn't in the sense it was an online sample,
> and in various ways this sample differed politically from true random
> samples": This claim was never made by me, in fact in the results paper for
>
> this study in the methods section, it reads "the results should not be
> taken as indicative of population-wide voter representativeness or
> statistically representative support for alternative voting systems." I am
> of an average mathematical mind, but I know better than that.

--uh, sorry if I misrepresented your claims; it was not intentional.
I believe I had read something (not written by you, i.e. not
written by Robin Quirke) claiming the online sample
was not just merely (a) the first 600 people who came along,
but rather (b) was intentionally chosen from a larger online pool via
an automated non-uniform sampling method intended to produce
a sample with more-representative demographics than (a).

While I applaud that idea, I claim (and Quirke seems to agree with me)
that it was not completely adequate. Which is not necessarily anybody's
fault; it just is not easy to do that, maybe not possible at all.
That is because people who self-select to be part of
online polls, are just not the same as random people, even
if you try to make them have similar racial, gender and/or age
composition. E.g. latino men age 30-40 who sign themselves up
for online poll duty, are not the same as latino men age 30-40
just generally. And we could give evidence for this failure.