[EM] More non-altruistic attacks on IRV usage.

[Kristofer Munsterhjelm]

matt welland wrote:
> On Sat, 2011-11-26 at 22:31 -0500, robert bristow-johnson wrote:
>> On 11/26/11 6:58 PM, matt welland wrote:
> 
>>> Also, do folks generally see approval as better than or worse than IRV?
>> they don't know anything about Approval (or Score or Borda or Bucklin or 
>> Condorcet) despite some effort by me to illustrate it regarding the 
>> state senate race in our county.
> 
> I wasn't clear. I want to hear opinions from the list: Is approval
> better or worse than IRV and why?

In my opinion, Approval is somewhere between IRV and "the advanced 
methods" (good Condorcet methods, MJ, etc).

The reason I think Approval is better than IRV is that while IRV makes 
its own decision about essentially whether to emulate people voting both 
Nader and Gore, or Nader alone, Approval lets the voters decide on their 
own. The voters can therefore approve both if it's more important to 
beat Bush than to support Nader over Gore, or approve Nader only if 
Nader's got a chance.

The reason I think the advanced methods are better than Approval is that 
they take this burden off the voters when the voters are sincere. If you 
vote Nader > Gore > Bush in Schulze (say), then you're both helping 
Nader to win against (Gore, Bush) and Gore to win against Bush. If Gore 
is a CW with a sufficient margin that you don't create a cycle - well, 
then Gore wins. Same with Nader.

If there's a cycle, it gets a bit more tricky. The method is easier 
influenced by strategy and your vote could hurt you. The Condorcet 
criterion no longer says what the answer should be, and the method thus 
has to use more indirect reasoning to find out who should win.

At least it narrows down the region in which strange things can happen. 
The good Condorcet methods pass criteria like Smith and independence of 
Smith-dominated alternatives, and so further narrow down these regions.

So, in short: IRV makes a guess as to which comparisons are the most 
important (using the logic of "least first-place votes = worst"), and 
when it gets it wrong, there's your center squeeze. Approval gives the 
decision to the voters, who will do better if they have access to 
polling data. Condorcet looks at more comparisons at once, while MJ 
reads ratings using robust statistics to satisfy criteria like Majority 
and to deter strategy.