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

[Jonathan Lundell]

On Dec 1, 2008, at 10:31 AM, Abd ul-Rahman Lomax wrote:

> At 12:54 AM 11/26/2008, Jonathan Lundell wrote: [attribution  
> corrected]
>
>>
>> My own view is first that we're talking about marginal differences
>> here, and that PR vs single-winner elections is of much, much greater
>> interest, and second that the interesting difference between   
>> plurality, IRV and other ranked methods is not in how they count  
>> any  particular profile, but rather in how they influence candidate  
>> and  voter behavior. In the IRV examples that Greg and Abd adduce,  
>> we don't  actually know what the ballots would have looked like if  
>> the elections  had used plurality. The set of candidates might well  
>> have been  different, the nature of the campaigns different, and  
>> voter strategies  different.
>
> That's correct. We can make some reasonable assumptions, though. We  
> can look at Plurality elections and look at how many voters vote for  
> minor candidates with no hope of winning. We can then look at IRV  
> results and convert comparable percentages of these votes to  
> plurality votes for frontrunners. I think we could get pretty close.

That, it seems to me, requires a leap of faith, or at least making  
assumptions that are hard to find objective verification for.

> Campaigns, from experience, don't seem to be much different. Voter  
> turnout doesn't seem to be much improved with IRV, though,  
> theoretically, it should improve turnout a little.

An exception is the second round of a TTR election. San Francisco is a  
case in point, and I'm guessing fairly typical, in having very poor  
turnout for their second round, less than half the first round, IIRC.  
This would be an advantage of any single-round method over TTR, of  
course.

> The fact, however, that the sincere preferences expressed by voters  
> for minor candidates don't shift results, in nearly all cases, acts  
> contrary to this effect, in the long run. So you get to vote  
> "sincerely," but with no net effect? You can "vote" more sincerely  
> by giving a campaign donation to a minor candidate!

My sense is that minor parties are optimistic that a) they'd  
demonstrate a higher first-round vote with IRV than plurality in any  
given election, and b) that the improved showing would have a  
beneficial effect on future elections. That is, the benefit of IRV  
(and again this wouldn't be exclusive to IRV) extends beyond any  
single election.

Is that true? I don't know, but there's some reason to think so. Green  
Party candidates running "real" campaigns have typically been able to  
obtain about 10% of the vote in California state assembly elections in  
recent years. Suppose they could reach 15-20% of the IRV first choice.  
That would give major-party candidates significantly more incentive to  
appeal to those Green voters for their second choice.

> Voter strategies in Plurality are based on main considerations:
>
> (1) Vote for the favorite. This actually works, it's better than one  
> might think. It is *highly* likely to produce the same winner as a  
> more advanced method, and it is only under special conditions that  
> it fails. So a lot of voters do this. Note that, by definition, most  
> voters, voting this way, are voting for a frontrunner, so this  
> "sincere" strategy is the same as consideration 2.
>
> (2) Vote for the preferred frontrunner, because all other votes are  
> moot. This is only important for minor party supporters who want to  
> express that for some other purpose than winning an election, such  
> as ballot position. Or for purely personal reasons, such as being  
> pissed at a party for not nominating their favorite, who is running,  
> and they either think that the possibility of the candidate from the  
> opposing part won't win in any case, or they don't care. Damn them  
> all! Yes, some voters vote like this, God bless them. It's part of  
> the system. Don't unnecessarily piss people off, make sure they see  
> the process as fair, if possible. Not always possible.
>
> Which of these strategies the voter actually uses depends on  
> preference strength. Tricky to estimate, but, again, I think that by  
> studying average behavior, we could get close.
>
> Essentially, we can look at IRV results and predict, probably within  
> a few percent, what the Plurality vote would have been. Most voters  
> probably use strategy 2 in Plurality; Plurality also tends to  
> suppress minor party candidacies which, with Plurality, show far  
> lower than the real party support, because of the strategic voting.  
> One of the benefits of Range, Warren calls it the "incubator  
> effect," is that a measure of third party support becomes available,  
> and there is little reason to assume that this would be insincere.  
> Why bother voting an "insincere" rating for a non-viable candidate?  
> Given that the vote has about zero chance of influencing the  
> outcome, in most elections.
>
> (But if you prefer a major candidate, it could be foolish, with  
> Range, to max rate a minor candidate. It doesn't accurately  
> represent your preferences, but for no good strategic purpose. Only  
> if you fear that the minor candidate might be the runner-up, and  
> lose to someone worse, would you do that, and this means that you  
> don't think this is a minor candidate. This might be a winner. I've  
> said this many times: if you would seriously regret a Range Vote,  
> under any reasonable scenario, including some election surprise --  
> like Le Pen getting second place -- don't vote that way. Adjust your  
> vote so it is safer. The Le Pen result was not expected, to be sure,  
> but those French elections, with many candidates, were very  
> susceptible to small variations. Had it been a Range election in the  
> primary, Le Pen wouldn't have made it to the runoff, I'm sure.  
> *Maybe* Range would have detected Jospin, maybe not. With "sincere"  
> Range votes, sure. But the Range winner, with no runoff, would  
> almost certainly have been Jospin. Note that IRV could easily have  
> missed Jospin, for the same reason that the plurality primary missed  
> him. Jospin would have, with even higher certainty, in a majority - 
> required Range election, been in a runoff against, again with high  
> certainty, with Chirac, and would have won.)
>
> (FairVote, to criticize Range, posits a truly preposterous election  
> scenario, where voters prefer one candidate over another, and 99%  
> vote 100 for the favorite and 99 for the other. A very small set of  
> voters (1 or 1% in the example) vote 0 for the first and 100 for the  
> second, and, of course, this outweighs the preference of a huge  
> majority. But it is an extraordinarily weak preference, and if it  
> was accurate, all voters will be very satisfied with this outcome,  
> and they will not be saying "damn! I wish I'd voted like that  
> selfish voter!" They will be, instead, if this is a real election,  
> and there were other candidates who might have won, with  
> significantly lower preference, very happy with the outcome and they  
> will all go out and party together. In other words, FairVote wants  
> us to think that the violation of the Majority Criterion is a Bad  
> Thing, when, in fact, the outcome is very good, better than I have  
> *ever* seen in a public election. If that majority would regret its  
> supposedly "sincere" vote, why in the world did they vote that way,  
> effectively abstaining from an important pairwise election. Surely  
> the exaggerated their real preference toward the top. That is an  
> error in voting, and errors in voting can always produce poor  
> results! In a real election, how likely is this error?)
>
> (We have 99% of the voters apparently thinking that there is a risk  
> that their favorite will lose, not to the second candidate, the 99,  
> but to someone much worse. However, in fact, there are *no* voters  
> who prefer the third candidate. If there is no third candidate --  
> FairVote doesn't specify -- then what we have is a situation where  
> the 99% majority essentially said, well, we have a tiny preference  
> for one candidate, but, really, don't listen to us unless the rest  
> of you really don't care. So they voted with votes of 1/100 vote  
> value in a binary election. Why? Only if they really don't care,  
> this is *almost* like a voter abstaining. Indeed, it's an almost  
> total abstention on the part of the voters. This vote is based on  
> the idea that voters are supposed to vote sincere absolute  
> utilities, though in this case, the utilities were normalized at the  
> top but not at the bottom. So the 99% may have been saying, I have a  
> *significant* preference for A over B, but because both A and B are  
> not Genghis Khan or Adolf Hitler, I'd have to be pretty grateful for  
> the election of either, so, to be honest with you, my preference for  
> A over B is only 1% -- tiny -- of my preference for B over C. I'd  
> die to prevent C from being elected. C isn't on the ballot? That's  
> irrelevant, I'm voting my "sincere utilities" because someone told  
> me this was better. And if they believed that, I can get rich, I can  
> sell them deeds to famous bridges. Voting is a choice, not a  
> sentiment. If this is a binary election, you are being asked to make  
> a "sincere choice," not make a "sincere rating." You can rate  
> "sincerely" if want, but if you don't want to waste your vote, you  
> will have to normalize your vote to the candidate set at the bottom,  
> too. Which would imply voting min for B, if there are only two  
> candidate. Voters *can* vote as described, but, then, it *clearly*  
> is based on a desire to abstain. These are votes, expressed in  
> fractions of a vote, multiples of 1/100 vote, they are *not*  
> sentiments. If you care about the outcome, show it!)
>