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

[Kristofer Munsterhjelm]

Abd ul-Rahman Lomax wrote:
>> Still, your point is taken. The problem with ordinal methods is that 
>> you can't specify strength; but that's also, in some sense, an 
>> advantage, since that means the method is less prone to being tricked 
>> by noise or by optimization. Which matters more is a question of 
>> perspective and what you want out of the method.
> 
> Sure. However, Range methods are actually less vulnerable to being 
> tricked by noise. Range votes are sincere votes, in what they express as 
> to ranking. Equal ranking is an important option, when equal ranking 
> isn't an option, then low-preference votes get just as much strength as 
> high preference ones. That's noise, and it can be severe. Those 
> decisions may be made last-minute based on sound bites on last might's 
> TV broadcast, an automated phone call received, etc. These phenomena 
> don't generally shift heavy preference, but they will shift minor 
> preference.
> 
> In Range, it's possible that the effect is damped, negative publicity, 
> not well-considered, may lower ratings some but it would have to be 
> pretty bad to shift it from 100% to 0%.
> 
> Methods which allow equal ranking devolve to Approval, which is a Range 
> method, when voters fully truncate.
> 
> Optimization and the similar normalization -- the latter being probably 
> almost universal -- do distort Range results, but only toward Approval 
> results. Which distort towards Plurality in the extreme. We fix that 
> with runoffs! Those extremes won't happen, apparently, and it seems that 
> even a few voters voting intermediate ratings in Range can beneficially 
> affect the result, can make it even better than either purely strategic 
> ("Approval") Range or fully accurate representation of preference 
> strength Range. More work needs to be done in the simulations.

Say that Approval distorts towards Plurality. What does Condorcet 
distort towards -- Borda? "Let's bury the suckers"? If people are 
strategic and do a lot of such distortion, wouldn't a runoff between 
Condorcet (or CWP, if you like cardinal ballots) and something resistant 
to Burial (like one of the methods by Benham, or some future method), be 
better than the TTR which would be the result of Approval-to-Plurality 
distortion? If people stop burying, the first winner (of the "handle 
sincere votes well" method) will become more relevant; if they don't, 
the latter (strategy resistant Condorcet) will still be better than 
Plurality, I think.

Now, that's probably a very complex system: first you have to define 
both the sincere-good and the strategy-resistant method, then you have 
to set it up to handle the runoff too. But it's not obvious how to be 
selfish in CWP (except burying), whereas it's rather easy in Range 
(->Approval, or to semi-Plurality based on whatever possibly inaccurate 
polls tell you). This, in itself, may produce an incentive to optimize. 
"I can get off with it, and I know how to maximize my vote, so why 
shouldn't I?"; and then you get the worsening that's shown in Warren's 
BR charts (where all methods do better with sincere votes than strategic 
ones). In the worst case, the result might be SNTV-like widespread vote 
management.