[EM] IRV vs Plurality ( Kristofer Munsterhjelm )

[Abd ul-Rahman Lomax]

At 01:13 PM 1/20/2010, Abd ul-Rahman Lomax wrote:
>At 09:09 AM 1/17/2010, Chris Benham wrote:
>>Kristofer Munsterhjelm wrote (17 Jan 2010):
>>
>>"To me, it seems that the method becomes Approval-like when (number of
>>graduations) is less than (number of candidates). When that is the case,
>>you *have* to rate some candidates equal, unless you opt not to rate
>>them at all.

Sure. However, in practical methods in the near future, not rating 
them at all is equivalent to bottom-rating. And all methods on the 
table equal-rank candidates not ranked.

In my view, preventing voters from equal ranking is preventing voters 
from expressing valuable information. However, not allowing them to 
rank when they have a significant preference also conceals valuable 
information. There is a compromise, such that information that the 
voter cannot express is of minimal significance.

>>That won't make much of a difference when the number of candidates is
>>huge (100 or so), but then, rating 100 candidates would be a pain.

Unless I can just lump them as I like. Bucklin, one of the early 
elections, had something like 100 candidates. People got excited 
about a method that would allow sincere voting without a spoiler 
effect.... and the method apparently worked. Three-rank ballot, 
probably. That means four ranks when we count bottom. Very simple to 
vote on 100 candidates.

1. Vote for favorite first rank.
2. Vote for any others first rank if you have difficulty deciding 
between them, i.e., what is really being done is defining a favorite 
*class* of candidates roughly equal in value.
3. Vote for any candidates you'd be happy to see election (not just 
merely not unhappy), but not favored, in second rank.
4. Vote for any candidates you would prefer to a runoff being held, 
in the last rank. Any rank may also be empty, and that conveys 
information about preference strength.
5. Don't recognize the candidate? Probably you don't vote for the 
candidate. Don't like the candidate, wouldn't want your vote to avoid 
a runoff? Don't vote for the candidate.

A rough equivalent of this ballot and voting approach would be 
Range4, with the value of 1 missing, you can't vote it. And range 
analysis could then be done. If the value of 1 is inserted as a 
"Disapproved but better than worst" category, then you'd get true 
range analysis, and if there is a significant difference between the 
Range winner and the Bucklin winner, I'd suggest a runoff might be held.

Contrary to the immediate reaction of many voting systems students, 
the Range winner, in a real runoff, if the data is accurate, has a 
natural advantage. It's because of differential turnout.

>>  I'd
>>say it would be better to just have plain yes/no Approval for a "first
>>round", then pick the 5-10 most approved for a second round (using
>>Range, Condorcet, whatever). Or use minmax approval or PAV or somesuch,
>>as long as it homes in on the likely winners of a full vote."

While 5-10 for the next round sounds good, it normally would also 
eliminated one of the major advantages of runoff voting: closer 
examination of the candidates remaining.

*If the top two are well chosen*, it's beneficial to have only two on 
the ballot. However, there are other possibilities. As an example, 
the candidates could be listed on the ballot in order of range 
ranking, so voters have the information from the previous ballot to 
guide them and may then vote as strategically as they like. Strategic 
voting in range is an expression of sincere preference strength in 
relevant races, that's been overlooked. It never involves preference reversal.

If somehow the wrong candidate makes it into the top two (quite 
unlikely with a good choice algorithm, though the possibility 
certainly increases with many, many candidates), write-in votes can 
be allowed in the runoff. If there is a missing candidate with 
significant preference strength, a write-in can win a runoff, and 
it's happened. It won't happen unless there is sincere and 
significant preference strength, and the vast majority of the time, 
the write-ins will be irrelevant, and if the runoff method is 
spoiler-free, like Bucklin as a simple example, no problem. Vote for 
your write-in and still cast an effective vote in the real election.

>>Simply using plain Approval to reduce the field to the top x point scorers
>>who then compete in the final round seems unsatifactory to me because
>>of the  "Rich Party" incentive (clone problem) for parties to field 
>>x candidates;
>>and because of the tempting Push-over ("turkey raising") strategy incentive.

Sure. However, Rich Party and Turkey Raising don't work with approval 
methods, beyond the communication advantages rich parties may have 
through greater media access. Bucklin seems to me to be a 
much-neglected approval method, with real implementation history, 
that behaves like approval, and whenever an election is close with 
more than two candidates, all ranks will be added in, so it *is* 
approval; but the adding in is sequential until a majority is found 
or until termination.

Which leads to a new method: Range/Bucklin. Range ballot, the ballot 
is essentially an instructed proxy to vote for the voter in a 
repeated series of approval election wherein the voter controls the 
approval cutoff inclusion point for each candidate, simulating what 
actually happens in repeated voting, the voters make compromises of 
increasing depth until some satisfactory result is obtained.

The proxies start out by voting for all candidates at the highest 
rank. If a majority, done. If not a majority, lower the approval 
cutoff one click. Iterate until a minimum approval threshold is reached.

For a deterministic method, the minimum approval threshold is any 
rating above zero. Or some higher value.

For a method requiring a majority, the minimum approval threshold 
would be midrange, or perhaps a little lower. Not much lower!

But the ballots can also be analyzed as pure range, i.e., sum of votes.

And if the range winner differs from the Bucklin (approval) winner, 
that's two possible runoff candidates. A Condorcet winner would be 
another, or the highest-rated or highest-approved member of a 
Condorcet cycle that doesn't include the previous two candidates.

In real repeated balloting, there are no eliminations at all. 
Eliminations are only provisional if write-ins are allowed, and they 
are a compromise in the name of efficiency and the value of improved 
voter attention.

My sense is that if the method chosen for the primary and runoff is 
good enough, loss of social utility through constricting the number 
of polls would be very small. And sometimes good enough is good enough.