[EM] Why I think IRV isn't a serious alternative 2

[Kristofer Munsterhjelm]

Abd ul-Rahman Lomax wrote:
> At 03:36 AM 12/25/2008, Kristofer Munsterhjelm wrote:
> 
>> Do you think my runoff idea could work, or is it too complex?
> 
> For years, attempts were made to find a majority using advanced voting 
> methods: in the U.S., Bucklin was claimed to do that, as it is currently 
> being claimed for IRV. (Bucklin, in fact, may do it a little better than 
> IRV, if voters vote similarly; my sense is that, in the U.S. at least, 
> voters will respond to a three-rank Bucklin ballot much the same as an 
> IRV one, so I consider it reasonable to look at analyzing IRV results by 
> using the Bucklin method to count the ballots. If the assumption holds 
> -- and prior experience with Bucklin seems to confirm it, Bucklin 
> detects the majority support that is hidden under votes for the 
> last-round candidates.
> 
> The methods generally failed; holding an actual runoff came to be seen 
> as a more advanced reform, worth the cost. All this seems to have been 
> forgotten in the current debate. Bucklin, IRV, at least one Condorcet 
> method (Nansen's, apparently) have been used.
> 
> The error was in imagining that a single ballot could accomplish what 
> takes two or more ballots. Even two ballots is a compromise, though, 
> under the right conditions -- better primary methods -- not much of one.

I think I understand your position now. Tell me if this is wrong: you 
consider the iterative process of an assembly the gold standard, as it 
were, so you say that all methods must involve some sort of feedback 
within the method, because that is required to converge towards a good 
choice. That feedback may be from one round to another, as with TTR, or 
through external channels like polls, as with the "mutual optimization" 
of Range.

Is that right?

> When considering replacing Bucklin or IRV with top two runoff, what 
> should have been done would have been keeping the majority requirement. 
> This is actually what voters in San Jose (1998) and San Francisco (2002) 
> were promised, but, in fact, the San Francisco proposition actually 
> struck the majority requirement from the code. Promise them majority but 
> given them a plurality.
> 
> If the methods hadn't been sold in the first place as being runoff 
> replacements, we might have them still! The big argument against 
> Bucklin, we've been told by FairVote (I don't necessarily trust it) was 
> that it did not usually find a majority. There is little data for 
> comparison, but I do know that quite a few Bucklin elections that didn't 
> find a first preference majority *did* find a second or third round one, 
> but in the long Alabama party primary series, apparently there was 
> eventually little usage of the additional ranks. But even the 11% usage 
> that existed could have been enough to allow the primary to find a 
> compromise winner. What they should have done, in fact, was to require a 
> runoff, just like they actually did, but continue to use a Bucklin 
> ballot to try to find a majority. This would avoid, in my estimation, up 
> to half of the runoffs. Since Bucklin is cheap to count and quite easy 
> to vote, this would have been better than tossing preferential voting 
> entirely.

If you're going to use Bucklin, you've already gone preferential. 
Bucklin isn't all that impressive, though, neither by criteria nor by 
Yee. So why not find a better method, like most Condorcet methods? If 
you want it to reduce appropriately to Approval, you could have an 
"Approval criterion", like this:

If each voter has some set X he prefers to all the others, but are 
indifferent to the members among X, there should be a way for him to 
express this so that if this is true for all voters, the result of the 
expressed votes is the same as if one had run an approval election where 
each voter approved of his X-set.

All methods that satisfy this will be limited to the criterion 
compliance of Approval itself, because criteria either pass or fail, and 
if it's possible to force the method into Approval-mode, then it's also 
possible to make the method fail any criteria that Approval does fail.

> Sure. Setting conditions for runoffs with a Condorcet method seems like 
> a good idea to me. One basic possibility would be simple: A majority of 
> voters should *approve* the winner. This is done by any of various 
> devices; there could be a dummy candidate who is called "Approved." To 
> indicate approval, this candidate would be ranked appropriately, all 
> higher ranked candidates would be consider to get a vote for the 
> purposes of determining a majority.

So, an approval cutoff. For a sincere vote, what does "approved" mean 
here? Is it subject to the same sort of ill definition (or in your 
opinion, "non-unique nature") that a sincere vote for straightforwards 
Approval has?

> In Range, it could be pretty simple and could create a bit more accuracy 
> in voting: consider a rating of midrange or higher to be approval. This 
> doesn't directly affect the winner, except that it can trigger a runoff. 
> Not ranking or rating sufficient candidates as approved can cause a need 
> for a runoff. If voters prefer than to taking steps to find a decent 
> compromise in the first ballot, *this should be their sovereign right.*
> 
> A Range ballot can be used for Condorcet analysis. Given the Range 
> ballot, though, and that Range would tie very rarely, it seems 
> reasonable to use highest Range rating in the Smith set, if there is a 
> cycle, to resolve the cycle. 

Hm, this may work, or at least be better than Range. Since the cardinal 
ballot is interpreted as an ordinal ballot - by rank as well as by value 
- there's not as great an incentive to compress-compromise. That still 
doesn't explain what the ratings of a cardinal ballot actually mean, 
though, but inasfar as people have an intuitive sense of what they do, 
it might work.

> Thus we'd have these conditions for a runoff:
> 
> (1) Majority failure, the Range winner is a Condorcet winner. (probably 
> the most common). Top two runoff, the top two range sums.
> 
> (2) Majority failure, the Range winner is not a Condorcet winner. TTR, 
> Range and Condorcet winner (cycles resolved using range sum).
> 
> (3) Majority, both Condorcet and Range, but Range winner differs from 
> Condorcet winner. same result as (2).
> 
> (4) Majority for Range winner, not for Condorcet. or the reverse. I'm 
> not sure what to do about this, it might be the same, or the majority 
> winner might be chosen. A little study would, I think, come up with the 
> best solution.

I think the easiest way would be to drop the probing of the majority. 
Just have a two-candidate runoff between the CW candidate and the Range 
candidate. If the two are the same, the second place would be the second 
candidate of the social ordering of either the Condorcet method or Range 
(unsure which). If there's no CW, discover it by the tiebreaking system, 
or if that's too complex, by Range.

There's also the somewhat strategy resistant variant that has been 
proposed earlier: voters input ballots that rank some or all candidates. 
All ranked candidates are considered "approved". Break Condorcet cycles 
by most "approved" candidate (or devise something with approval 
opposition to preserve clone independence, etc). The point, at least as 
far as I understood it, is that you can't bury without giving the 
candidates you're burying "approval", thus burial is weakened.

> Range is theoretically optimal, as optimal as is possible given an 
> assumption that most voters will vote a full strength vote in some pair. 
> However, normalization or poor strategy can result in distortion of the 
> Range votes compared to actual voter utilities. One of the symptoms of 
> this might be Condorcet failure for the Range winner. If it is true that 
> the Range winner truly is best, then we have a situation where the first 
> preference of a majority might not be the Range winner, or, supposedly, 
> the Range winner vs the Condorcet winner might award the election to the 
> Condorcet winner. But, in fact, it is normal for small electorates to 
> set aside the first preference of a majority in favor of some greater 
> good. I see no reason not to extend that as a possibility to large 
> electorates. My own opinion is that, *if the Range votes are accurate*, 
> the Range winner will normally beat the Condorcet winner, because of 
> differential turnout and weak preferences that reverse during the runoff 
> campaign. On the other hand, having this runoff possibility answers the 
> common objection to Range: majority criterion failure.
> 
> With a runoff system like this, MC failure doesn't occur, because the 
> majority in the final and effective election has voted for the winner.
> 
> (That's not 100% absolute, if write-in votes are allowed in the runoff, 
> as I believe they should. But if the runoff is, say, Bucklin, maybe 
> two-rank, I think that spoiled majorities will be relatively rare, and 
> that the elections that end up with a plurality would almost always be 
> resolved the same way if an additional runoff were held. Thus, short of 
> Asset Voting, this could be almost perfect.

You could also use the same method as in the first round, which saves 
you the trouble of having to define it twice. Just show that it reduces 
to "elect whoever has the most votes" in the number of candidates = 2 
situation - that's the only strategyproof method when there are only two 
candidates.

> Want perfect? Asset Voting, which bypasses the whole election method 
> mess! Single-vote ballot works fine! And that's what many or even most 
> voters know how to do best.

Or have a parliament and bypass the whole thing.

> So I've shifted to proposing Bucklin, though Approval remains a simple, 
> do-no-harm, cost-free reform. Introduce it to a TTR system, some runoffs 
> may be avoided. Introduce Bucklin, more.

Do you propose Bucklin because it gradually transforms to Approval (as 
more ranks are counted), or because of its own properties? If the 
former, the criterion I gave above might be useful if we can find a 
method that is better on its own terms yet has that property. If the 
latter, I think other methods are better. I'll note, though, that it's 
quite easy to make a PR version of Bucklin (and I've done so in an 
earlier post), so that claimed advantage of IRV would also hold for Bucklin.

> But Range Voting, a ranked form, was written into law in the U.S., I 
> think it was about 1915. Dove v. Oglesby was the case, it's findable on 
> the net. Lower ranked votes were assigned fractional values; I think it 
> was 1/2 and 1/3. Relatively speaking, this would encourage additional 
> ranking, I'd expect.

By that reasoning, any and all weighted positional systems are Range. 
Borda is Range with (n-1, n-2, n-3 ... 0). Plurality is Range with (1, 
0, 0, ... 0). Antiplurality is Range with (1, 1, 1, ..., 0), and so on.