Response to Christopher Billman
28 views
Skip to first unread message
Clay Shentrup
Jan 7, 2017, 12:18:58 AM1/7/17
to The Center for Election Science, Christopher Billman
https://docs.google.com/document/d/1Xe70JCCfadyI4S_QhOfm6Y51tLSiaXa9arQS75wzmlo/edit
> Monotonicity is also not a concern to me because it’s effect is impossible to see before an election, therefore it is impossible to use to alter the election, and its pure existence is basically meaningless.
This is a common misunderstanding. The problem with monotonicity has nothing to do with tactical exploitation. The problem is that it means you have a mathematical proof that the wrong candidate was elected in at least one of the two "mirror opposite" scenarios. I explain this in simple terms here.
> In order for monotonicity to occur at ALL, there have to be very specific conditions already in place in the electorate.
Well, it has happened. E.g. in Burlington, VT and Frome, Australia.
And these kinds of IRV paradoxes tend to be more common in elections where IRV produces a different outcome than Plurality Voting would have.
> I agree, but in this example those voters are ALSO moving towards red
Maybe in the example you're talking about, but it can happen regardless. As you can see in the example in my video, the electorate "gets closer to" (increases their support for X) without changing their opinion for any other candidate—and that alone causes X to go from winner to loser.
> IRV, even with monotonicity occurrence will result in the winning of a candidate who had a great deal of support originally to begin with, and who, by the end, has a majority (50% +1) of support.
I'm afraid that's incorrect. See this example:
IRV only guarantees that the winner is preferred by a majority to at least one other candidate. He could be opposed by a majority vs. all others.
Money quote:
Instant Runoff Voting selects candidate X as the winner, beating W in the final round, 65% to 35%.
But wait!
A huge 67% majority of voters would rather have candidate Y than X. And Y received nearly twice as many first-place votes as X, 32% vs. 17%.
And an even larger 83% super-majority of voters would rather have candidate Z than X (and Z got just a little fewer first-place votes than X).
> On the flip side of monotonicity there is something called later no harm, which I will cover momentarily.
Later-no-harm is an "anti-criterion". Satisfying it is harmful.
You want the Favorite Betrayal Criterion.
Or as Warren recently pointed out in a very clever table:
For short, let "A hurts B" mean (in some voting system)
"there exists an election where by indicating your
Ath choice, you hurt your Bth choice (and asterisk:
if you prefer A over B then 'A winning' does
not count as 'hurting' B, it is just 'the natural course of events')."
problem IRV approval
1 hurts 1? yes no
1 hurts 2? yes no
2 hurts 1? no yes
2 hurts 2? yes no
> Majority support is very important to IRV, which is why IRV is set up to promote it.
There's no clear definition of "majority support".
> With AV [Approval Voting], winners can (And have on multiple occasions) happen with less than 30% of the vote “supporting” them.
There's nothing wrong with that. This is simply another logical fallacy which I explain here.
Money quote:
Only 17% of voters approve of X [the IRV winner], which is far lower than Y's 49% approval. Richie complains that Approval Voting elects someone with only 49% approval—but his IRV system elects someone with only 17% approval.
X isn't really a "majority winner" after all! Note that Y is preferred to X by a huge 67% majority of the voters (35% + 32%). And Y is the first choice of 32% of the voters, compared to only 17% for X. Y is thus the true majority winner, not X. So Approval Voting, not IRV, elects the true majority winner here.
> simply research Dartmouth college alumni's attempt at using AV and what happened over the few years they used it
Warren addresses the Dartmouth rhetoric here.
Money quote:
So to eliminate the "threat" of "bullet voting," force everyone to bullet vote and discard their ballots if they do otherwise?!?! [I.e. revert to Plurality Voting]
> or Institute of Electrical and Electronics Engineers’ attempt in the 1980s.
Addressed here.
> To this I would simply point out that Australia hand counts their ballots . .. ALL of them. EVERY election.
A few points:
1) IRV is not precinct summable, so regardless of whether you hand count it or use machines, you cannot simply sum the precinct results to get a sum. You have to translate the ballots to a central location for counting (or "effectively" do so by getting each round's totals submitted to a central authority.)
2) IRV results in more spoiled ballots, whereas Approval Voting results in fewer.
3) IRV has a higher chance of tie/near-tie elections.
So you can hand count IRV of course, but it's more complex. That complexity has apparently incentivized (if if not forced) the use of more complex voting machines here in the San Francisco Bay Area where I live.
4) Most people I know in Berkeley and San Francisco cannot even explain how IRV works. See this conversation with a former coworker of mine who's a brilliant engineer on Adobe's TypeKit.
So the logistical concerns are certainly more than fear mongering.
> I have encountered several [flaws with Approval Voting], else I would be a big AV proponent.
I contend that what you call "flaws" in Approval Voting are not actually flaws, upon deeper inspection. E.g. Later-no-harm vs. Favorite Betrayal Criterion.
> [Warren's] work is based on a few presumptions including, majority rule is not important
A) There is no actual objective definition of "majority rule". Indeed, IRV can elect X even if Y is preferred to X by a majority of voters and got more first-place votes than X.
B) It's not a "presumption" that majority rule isn't important—it is mathematically proven.
> This [utility monsters] strikes me as VERY unpleasant.
But unfortunately necessary.
> AV/Score are seeking a candidate with broad support
Not broad support per se, but the greatest combination of breadth and depth of support. E.g. a candidate with two five's has the same value as a candidate with a zero and a 10. And indeed, this is the right goal. A rational person wants to maximize his expected welfare.
> where IRV is seeking a candidate with majority support.
A) There's no clear definition of "majority support" — indeed, all voters may "disapprove" of all candidates, but one of them is still going to win a head-to-head race "by a majority"
B) IRV can fail even the most basic definition of "majority support" to the extent it is even a meaningful term.
Reply all
Reply to author
Forward
0 new messages