[EM] IRV vs Plurality

[Abd ul-Rahman Lomax]

At 04:58 PM 1/10/2010, Jonathan Lundell wrote:
>On Jan 10, 2010, at 1:02 PM, Abd ul-Rahman Lomax wrote:
>
> > At 01:06 AM 1/10/2010, Jonathan Lundell wrote:
> >> On Jan 9, 2010, at 9:23 PM, Abd ul-Rahman Lomax wrote:
> >>
> >> > On the other hand, in one-third of nonpartisan top-two runoff 
> elections, which IRV supposedly simulates, the runner-up in the 
> primary goes on to win the runoff, a "comeback election," according 
> to a FairVote study. It simply does not happen with IRV.
> >>
> >> It's hard to know what to make of this claim, other than, "so 
> what?" -- if what you're comparing to TTR is the IRV candidates 
> with the highest first-choice ballots.
> >
> > I'm comparing *elections*. Yes, not all questions are addressed, 
> but the response of "so what?" misses the point, obviously. If we 
> assume sincere voting in TTR
>
>I would not. TTR encourages favorite betrayal for much the same 
>reason FPTP does.

"If we assume" is for purposes of consideration, not as a claim of 
the pure existence of fully sincere voting. Indeed, no method but 
Asset voting allows this purely without penalty.

My sense, from examining real votes in top two runoff elections, is 
that voters do generally vote for their favorite, under one of two 
scenarios: the candidate they vote for is truly their favorite, or 
the candidate is their favorite among those with any kind of chance 
of winning. What voters do with respect to other candidates, outside 
this, is largely moot, isn't it? Voters with top-two runoff will take 
chances and vote for a lesser candidate in the primary, much more 
than with simple Plurality, but less than with IRV or better methods.

> > and IRV, and if we consider real world conditions (separate 
> runoff, not merely some simulated runoff with the same voters and 
> the same preferences and no new knowledge or shift of preference or 
> better consideration, etc.), the results are quite different. That 
> preference order doesn't shift in nonpartisan IRV elections, though 
> the vote transfer process, was completely unanticipated, as far as 
> I know. Mr. Lundell, were you aware of it? Did anyone predict this?
> >
> > What this means that, with very little exception, you can use 
> Plurality instead of IRV.
>
>Again, I would not make that assumption. In this case, not simply 
>because of the difference in voter behavior and voting rule, but 
>because the nominating incentives are radically different.

But it is the whole picture that counts, not just one part in 
isolation. Here is what is apparent: if voters vote in Plurality as 
they would vote with IRV, but only for their first preference, they 
will get the same results as IRV, generally, if the election is 
nonpartisan. You cannot predict this from mere voting systems theory, 
it would take a study of voting psychology and mass political 
behavior to understand it, I suspect.

Now, will voters vote that way, if the method reverts to plurality? 
Other things being equal, I suspect so, except that they may be able 
to improve results over this, by paying more attention to the 
nominating process and by making better compromises. For example, 
sincere first preference will pass over a compromise candidate, with 
both plurality and IRV, but the problem is worse with IRV because it 
is not expected. Favorite Betrayal grates far more with IRV because 
of the pretense that you can just vote IRV sincerely and all will be well.

For this reason, the effect on nominations and voter behavior, I 
strongly suspect that Plurality -- in nonpartisan elections, again, 
please assume that context when I make these sweeping statements -- 
is actually better than IRV. Top-two runoff will have certain 
problems similar to IRV, but can make up for these in other ways with 
the increased voter awareness that takes place in the runoff, and if 
write-ins are allowed in the runoff, there are additional 
possibilities. (Write-in campaigns in runoff elections can and have 
been successful, and the riskiness of it can easily be eliminated by 
using Bucklin for the runoff, even if only two ranks.)

It is not rational to compare only two systems in considering reform, 
and voting systems are combinations of many details, and various 
combinations of details may work to improve a system over the general 
case for general implementations of that system.

For example, a rigid majority requirement is a detail, it has nothing 
to do with how the ballot operates and how the votes are canvassed, 
it is only a tacked-on requirement. But that requirement turns a very 
primitive system, plurality, vote for one, into a very powerful and 
quite sophisticated system, and the deviations from this are, in 
fact, due to the abandonment of the majority requirement in a runoff, 
and to candidate eliminations based on incomplete information, 
neither of which are permitted in standard deliberative process.

To compensate for that, if it is considered that an election must 
complete, then, the use of a more advanced ballot method in the 
primary and a single runoff, can much more closely simulate what 
would take place in a longer series of FPTP elections, and lead to 
results, with reasonable efficiency, that, at worst, involve very 
little Bayesian regret over the ideal minimum.

I know that I sound like a broken record sometimes, with Bucklin ... 
Bucklin ... Bucklin. But Bucklin is really Approval Voting with one 
simple tweak; ranking allowed. As I'd suggest it be implemented, it 
is Bucklin/ER. Bucklin as used in Duluth only allowed equal ranking 
in third rank, but there really isn't a sound reason to not allow it 
in all ranks, as far as I can see. So the voters may vote it as if it were:

Plurality
Approval
Instant Runoff Approval, with limited Later No Harm protection.

I believe, not only from my expectations but from historical 
experience, that voting Bucklin will be easily understood by voters, 
and that many of the objections to Bucklin have been based on 
misunderstandings. For example, in some historical Bucklin 
applications, the usage of additional ranks declined to not much more 
than ten percent. That is not a harm, particularly in party primary 
elections, which was the application. If the elections are political, 
final elections (not sure if this level of truncation happened in 
political elections), ten percent is quite enough to handle the 
normal spoiler effect. Note that voters who prefer one of the top two 
have no incentive or need to rank any other candidate, the voting 
system merely needs to accomodate those who have some other 
preference, and, as well, to handle the situation reasonably where 
there are three major candidates. IRV falls on its face in the 
three-major-candidate situation, being highly vulnerable to center 
squeeze (like simple TTR), but Bucklin will generally handle it well. 
Used as a primary election method in a system that requires a runoff 
if the winner doesn't get a certain threshold of votes (typically a 
majority, but there are other possibilities), Bucklin could be very 
close to ideal, as I've written, which is quite a bit to say about a 
system which is so easy to vote and canvass.

Much more than IRV, Bucklin allows a *completely* sincere first 
preference vote, one would only equal-rank first preference if the 
preference strength were small.