[EM] IRV vs Plurality

[Abd ul-Rahman Lomax]

At 09:47 AM 1/15/2010, Kathy Dopp wrote:
>Your steady stream of false claims about me in your
>recent emails show us much more about yourself than reveal anything
>about me.

To those with eyes, most everything we write reveals much about us. 
However, Kathy, I suggest you let others defend you; correct factual 
errors, if they matter, but don't get caught in the personal 
conflicts, behave as the professional you are. As an expert in your 
field, you will always encounter ignorant objections and attacks.

I think it was important for you to point out that you are 
non-partisan in your work, and I also don't believe that you are 
partisan on the issue of voting systems, beyond what your expertise 
has led you to understand, you are "partisan" on issues that make 
maintaining election integrity more difficult. And you have become, 
therefore, somewhat of an advocate of not adopting or of dropping such systems.

But that kind of partisanship is common with experts.

To apply this to a question that sometimes comes up, the issue of how 
important strong preferences are. Are strong preferences the result 
of expertise, or of fanaticism? Opponents of Range Voting, which 
respects expressed preference strength, often seem to assume that 
those with strong preferences are partisan fanatics. But is this a 
sane assumption, is it true, on average?

The same line of thinking causes people to assume that there is 
something wrong with low turnout in runoff elections, because those 
who vote are those who care enough to take the trouble to vote. I can 
assume that those who care are not necessarily a fair sample of the 
entire electorate, so, indeed, they may make decisions that would not 
be made by the entire electorate, if you forced the latter to vote. 
(They do this in Australia, actually.)

But which decisions are better for the society? By definition, those 
who don't care as much have less of a stake and less concern about 
the outcome, and, on average, they have less knowledge about the 
implications of each alternative.

One of those expressing complaint about low turnout in runoffs notes 
that it allegedly favors Republicans. If that were true, I'd hasten 
to check out the Republican party. It would be a sign that it was the 
party of those with more knowledge! But it's not generally true; in 
some circumstances, though, the Republicans may be better organized, 
and perhaps this is even due to better funding.

But the solution isn't to use voting systems that force everyone to 
vote, which merely makes results *even more* susceptible to 
manipulation by those with the most funding. The solution is to 
organize the people directly, for the people have resources that are 
even better than money, and more powerful, *if organized.*

Poor people can generally turn out to vote, and it's even possible 
that they could find it easier. If you are out of a job .... and if 
you have a supportive community that will provide you with 
transportation, babysitting, etc. ... you can vote. But if you are 
indifferent to the options, in fact, no amount of encouragement to 
vote will help.

There are voting systems that allow people to see the effect of their 
vote, even if they don't "win," but mostly they are not on the table. 
Good voting systems don't just determine winners, though that's the 
primary purpose. They also collect and provide accurate information 
as to public preferences, guiding future elections and campaigns, and 
because collecting that information is cheap if it's part of an 
election process, and if the information collected is the kind that 
would allow predictions of future voting behavior, a great deal of 
general social benefit can be generated quite aside from the benefit 
of making decisions.

And this is another reason why we should collect preference strength 
data, because raw preference, simple preference order, does not allow 
us to make predictions with high confidence. An A>B preference that 
is so small that tomorrow it might be B>A is quite different from a 
strong preference! -- which takes an earthquake, so to speak, to 
reverse. Suppose a candidate loses an election with a very poor 
showing in first preference votes (which may be the only votes on a 
plurality ballot). But suppose this candidate is not only every 
voter's second choice, but the preference strength of the voters of 
someone else over this candidate is low. With support and better 
campaigning, that compromise candidate would be quite likely to win a 
rematch. And would, indeed, quite possibly, be a much better choice, 
uniting the community.

Hence if you asked me about the best voting method, I'd propose a 
Range ballot, of moderate resolution. Range ballots can be analysed 
as preferential, ranked ballots, with equal ranking allowed. If there 
is an explicit approval cutoff, they can be used to provide Yes/No 
information, to determine if a majority actually supports an outcome, 
which is highly useful and fits in with common understanding of the 
importance of majority approval of a result. They can be used to 
determine a condorcet winner. (If the number of ratings is high 
enough, we can assume that equal ranking means equal preference, and 
that if the voter has a significant preference, it will be expressed, 
because high Range resolution corresponds to an ability to fully 
rank, and if the voter doesn't do this, they have not sacrificed any 
significant voting power for strategic purposes.)

A high-resolution Range ballot, though, complicates the voter's task, 
to some degree, but -- this is important -- no more than a fully 
ranked ballot. One trick would be to make the range resolution equal 
to the number of candidates, so if a voter wants to do it, the voter 
can simply rank the candidates, which for some might be easier. And 
that will approximate their actual sincere ratings, and if the 
approximation isn't accurate enough, then the voter simply shifts the 
ratings of some candidates, causing "overvoting" at some ratings 
(ranks), and empty ranks.

And then you can analyze such a ballot in quite a number of different 
ways. You could even use it for IRV! You could use it for any 
Condorcet method. You could use it for Plurality, if you wanted to. 
Further, voters who simply bullet vote (vote for one, full rating, 
and no rating or zero rating for all the others) have still provided 
ranking information, with all but one candidate equal-ranked bottom. 
That's actually often a sensible vote, and doesn't collect noise. 
When used in a top-two runoff situation, this is the meaning: "I 
don't know enough about all these other candidates to make a sensible 
choice between them. If a majority don't prefer my favorite, ask me 
again when I'll have better information!"

But other voters *will* have some preferences among those remaining, 
and could express them. And so we collect deeper information about 
the electorate.

Bucklin would take such a ballot and start by looking at the highest 
rating. Is there any candidate rated at the highest rating by a 
majority? If so, that's the winner! (with one possible exception I 
won't address here).

If no such candidate, then the ballots would be reanalyzed down some 
distance. With computers, you'd just knock the "Yes vote" down one 
notch, and look for a majority, and you'd keep doing this until you 
have a majority. If this is a deterministic election (say it is a 
runoff!), you keep clicking it down until you have a majority, or you 
have gone all the way down and have counted all the votes that were 
above zero. And then the winner is the one with the most votes. Every 
vote has been counted. How often would there still be majority failure?

Sometimes, I'd bet. Depends, and we don't really know. But this use 
of every ranked vote is important. Arguments with this Range/Bucklin 
method will come from Range advocates who would want to simply count 
every vote and not allow a majority to prevail, because they know 
that the Majority criterion is defective. But ... that's actually 
easy to fix. Look to see if the full-count Range winner is the same 
as the Majority winner at some higher approval cutoff. If so, done. 
If not, runoff. There is no way to tell, then, from the ballots, who 
will win the runoff. My bet, though, normally, would be on the Range winner!

And far more sophisticated analysis is possible. But if we start with 
majority approval or a runoff, reasonably defined, we can start to 
collect the data that will allow avoidance of most runoffs. And we 
will avoid, my guess, more than half of runoffs, without doing any 
violence to the principle of majority rule.

With a method that is precinct-summable, and that can start out 
*very* simple, either with simple Count All the Votes on the same 
ballot as plurality (but with slightly different instructions), and 
that can be made much more flexible and satisfactory with the 
addition of ranks to be counted if necessary, and those ranks can 
actually be used as range ratings.

Consider that voters are dividing candidates into four groups: 
Favorite, Preferred, Barely Acceptable, and No Vote. Barely 
Acceptable means "I'd prefer to see this candidate elected than see a 
runoff election." Very simple, very easy to understand, and very low 
election pathology, if any.

The optimal vote, or very close to it, would be fully sincere. The 
voter would not have to put *any* votes in any rank. If the voter is 
actually indifferent, but does have a preference among a set of bad 
candidates, the voter can put only votes into Barely Acceptable!

To make this into a full linear Range ballot, you would add another 
voting category: Disliked, and you might have an explicit No Vote 
category, which would be renamed as Worst. There should be no 
restriction on the voter, the voter may put as many candidates as 
desired into any of the categories, and would know that part of what 
would be done would be to analyze the votes as sums of category 
values, being 4, 3, 2, 1, 0. (I'll leave aside the question of 
individual candidate abstinence and will point out that it's possible 
to start with traditional counting, which treats No Vote the same as 
a No on the question of the election of the relevant candidate, and 
then collect data, and then, later, if it seems reasonable, to move 
to Average Vote for part of the method, with some required minimum 
explicit vote quota). With this, any vote of 2 or higher would be 
treated as a Yes, allowing determination of a majority if that's 
important -- as I believe it is.

Whenever the outcome is ambiguous, a runoff is possible. In a totally 
ideal system, there would be no eliminations in runoffs, voters would 
remain free, and a decent compromise is to put on the later ballots 
the two (or rarely more than two, as in Vermont gubernatorial 
process) most likely candidates to receive majority approval, with 
write-ins allowed and using a method that isn't subject to the 
spoiler effect. If this were truly no-elimination, as with standard 
elections in voluntary organizations under Robert's Rules, then the 
same ballot would continue to be used. To make this practical, we'd 
probably have to move to internet elections, and there is a way to do 
that safely. It's quite simple!

Use Asset Voting, and in the runoffs, all voting is public, by the 
Asset Electors, being the candidate(s) in first position on each 
ballot. The set of electors in subsequent elections is fully 
representative, through sincere voluntary choice, of the general 
electorate, all who voted in the first election. That ought to 
satisfy people who think runoff elections aren't fair!

There is a common belief among students of voting systems that no 
election method can be ideal, but this is partly based on a defective 
understanding of Arrow's Impossibility Theorem, as well as a 
too-common lack of understanding of the purpose of voting systems 
compounded by a lack of analysis of the overall process, which 
includes the effect of preference strength on turnout.

What I have laid out is not only a high-performance voting system, 
but one which can be implemented in baby steps, starting with Count 
All the Votes, then addressing a simple defect of that, the inability 
to express first preference when voting for more than one, by using 
ranking with a method which was widely used in the U.S. at one time, 
and with which there is a great deal of election experience.

It's been believe that Approval Voting hasn't been used in the U.S. 
But that's because the actual application was of a method that almost 
always defaulted to Approval Voting in actual practice, but that 
wasn't the simple one-rank form, it was more advanced. Students of 
Approval Voting have frequently studied it in repeated elections, 
where voters start out with high approval standards, then gradually 
lower them to add more approvals in order to find a compromise 
result, and this iterative process is known to be of high 
performance, and the only problem is the number of ballots it can take.

Bucklin voting, from a century ago, was precisely this, simulated 
with three rounds on a simple ranked ballot. The ballot only included 
"approved" rankings, i.e, the voter only voted in some rank for a 
candidate if the voter was willing to support the candidate as a compromise.

The major defect, in my opinion, was that, like IRV today, Bucklin 
was sold as a replacement for runoff voting, "find a majority with a 
single ballot," and, when it failed to do that in some elections, 
there was a backlash and the forces that didn't want better voting 
systems were able to fail. Don't oversell! Bucklin will reduce the 
number of runoff elections if used as a primary method, and it makes 
write-ins safe for runoffs. So Bucklin will definitely save money 
compared to pure vote-for-one runoff top two runoff, but won't suffer 
from center squeeze.

Count All the Votes.

Did I mention that we should count all the votes?