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

[Abd ul-Rahman Lomax]

At 06:14 PM 12/21/2008, James Gilmour wrote:
>Abd ul-Rahman Lomax   > Sent: Sunday, December 21, 2008 1:44 AM
> > LNH as an absolute principle, which, as an election criterion, it is,
> > is harmful.
>
>That is a value judgement  -  which of course you are perfectly 
>entitled to make.

Sure. The desirability of any criterion is generally a value 
judgement. However, there are fundamental values and less fundamental 
ones, consensus values and idiosyncratic ones, ignorant values and 
informed ones. What I've heard from you, James, is that LNH is 
desirable because, without it, some voters (how many? under what 
conditions?) will be inclined to not add additional preferences, 
fearing "harm" to their favorite. Apparently, hang the outcome!

I said "as an absolute principle." There is a political issue. If a 
voting system doesn't satisfy LNH, but the situations where that's 
possible, where that would actually impact the voter's interests, 
it's politically inconvenient. One can't say "Never." I've seen, 
however, that LNH has been asserted for a method where, in context, 
it simply wasn't true, it was possible for a later preference, with 
IRV, to "harm" the favorite's election prospect, because it was IRV 
where a true majority requirement was maintained. *LNH is 
incompatible with a majority requirement, unless the further process 
considers all eliminated candidates permanently eliminated, no matter 
what. It's incompatible with direct democratic process.*

When I wrote "absolute," I meant that some forms of LNH protection 
were quite possibly useful.

How, indeed, do we judge voting system criteria? Is there any 
approach that isn't subjective? How much harm is done by those 
alleged truncations because of the theoretical possibility of LNH 
failure. I've never seen any evidence that truncation is less common 
with IRV than it was with Bucklin, and there is some evidence to the 
contrary. Bucklin. Should I say, "the American Preferential Voting 
System," as it was called?

It was an Englishman, though, who noted the probable reason for most 
truncation, and it ain't LNH fears. It is a combination of ignorance 
regarding remaining candidates and, on the other hand, strong 
preference for the favorite and relative disinterest in who wins if 
the favorite doesn't win. Preference strength. Don't leave home 
without it. It makes all kinds of voter behavior far more understandable.

> > It prevents the system acting as a negotiator seeking
> > compromise, because it prevents compromise until and unless the
> > favorite is eliminated. Frankly, I doubt that anyone who fully
> > understands the implications would prefer an LNH system to one which
> > more appropriately negotiates on behalf of the voter, seeking the
> > best compromise. LNH means *no compromise unless you eliminate my
> > candidate totally!* That kind of position will readily be seen as
> > fanatic, intransigent, and selfish, in normal negotiation situations.
> > LNH in a system *enforces* this, requiring all voters to be just this
> > intransigent.
>
>I would hesitate to describe the electors I have experienced as 
>"fanatic, intransigent and selfish".

Of course not, but they are voting in a system that makes it 
unnecessary, if they are voting in an STV system. The system does it 
for them, and what I see from Mr. Gilmour is that this is what they 
want. I don't think they would really want it, if they understood the 
implications, and I suspect that he has not explained them to them.

I said that, in direct personal negotiations, this kind of behavior 
would generally be seen that way, not that voters were this way.

>   What interests me particularly
>is that their insistence on LNH (or at least, their reaction to the 
>effects of its presumed absence) is an intuitive response.

Sure. Like lots of intuitive responses, it's less than optimal. There 
is nothing wrong with the response, per se. Bullet voting as an 
initial stage in negotiation is perfectly normal and functional. 
"This is what I want," people will say. They don't say, if they have 
a significant preference, "This is what I want, but if you don't like 
that, this other option is fine with me." That's giving away the farm 
for a small price. So I expect some kind of reluctance to disclose 
lower preferences. The strongest effect would be with Approval, which 
only allows equal ranking top or bottom. Bucklin fixes that, though, 
without *enforcing* LNH under all circumstances. Voters will know how 
to use it.

Optional Preferential Voting, in Australia, sees massive truncation. 
Truncation is normal, when voters are free to do it. *Most voters* in 
*most public elections* will truncate, they won't use up all the 
ranks even in a 3-rank system.

>Other comments made by ordinary electors over the years lead me to 
>suspect that this intuitive response reflects the importance
>ordinary electors attach to their first preference.

Yes. It's functional, as I wrote above. The problem arises when the 
system *can't* move beyond that, until or unless the favorite has 
been eliminated. What if a voter, somehow, could choose LNH or not? 
Bucklin did allow some flexibility, I think, it was possible to leave 
the second rank blank, or to vote for a write-in or hopeless 
candidate (write-ins here would allow that vote to be sincere) in 
second rank, then write in the real fallback vote in third rank. How 
strongly does the voter want to insist on "over my candidate's dead 
body?" Basically, insuring LNH compliance requires that the favorite 
is unable to gather lower ranked votes from other non-eliminated 
candidates. It cuts both ways. It's entirely possible that, while my 
lower ranked vote isn't going to hurt my candidate, the system does 
it for me, it requires my candidates elimination before my lower 
votes (anyone's lower votes from non-eliminated candidates) can be revealed.

Tell that voter that because of LNH, their candidate may lose even 
though the candidate would be the LNH method's winner by a two to one 
margin, will they still be so pleased by LNH.

LNH sounds good to someone who doesn't understand how these systems 
behave. It doesn't sound so good to people who understand the range 
of possibilities. Limited protection for the favorite, absolutely. 
Good thing. But absolute protection, not so good. It's possible to do 
pairwise analysis such that the voter's lower ranked vote doesn't 
count in the pairwise election between the favorite and the less 
preferred one. But this leads to further complications, and it's not 
clear that it actually improves result. I do know that the LNH 
compliant IRV can easily show poor results, that's what the 
simulations show, and that non-LNH methods do better. What does the 
voter prefer, LNH compliance or better results? (Which means, on 
average, better results from the voter's point of view. Better 
results raise all boats, on average.)

Do these voters understand that their lower ranked vote in a method 
like American PV doesn't actually "harm" the favorite. It helps the 
second ranked candidate win over all others, but not over the 
favorite. The net result of this, in American PV, is that the voter 
stood aside and said, effectively, okay, I'll let the rest of you 
choose between my favorite and my second best. I'll support either 
one, in the end. If the voter doesn't want to do that, the voter 
should bullet vote, plain and simple. Many will anyway.

Bucklin doesn't prevent intransigence, it merely doesn't *enforce* it.

>   I know some voting system theoreticians say that no more 
> importance should be
>attached to a first preference than to any other preference, but I 
>don't think ordinary electors view the world that way.

I don't think that, and no Range theorist thinks that, and it's 
blatantly preposterous. The first preference, if it's not a weak one, 
has very real meaning. I support allowing equal ranking in first rank 
with Bucklin to cover the situation where the voter really doesn't 
have significant preference between the top two for the voter. Pure 
ranked methods don't allow that, forcing, then, the expression of 
nonexistent or trivial preferences. Noise. Clearly this can damage 
overall social utility. And the voter does not benefit from that 
restriction, from mandatory ranking.

>   And as a
>PRACTICAL reformer, it is ordinary electors who concern me (along 
>with the politicians and party activists we have to get on-side if
>we want to achieve actual reform).

Sure. However, there are lots of considerations that affect the 
ordinary electors. My opinion is that informed electors won't pick 
English Preferential Voting if they understand the differences 
between it and American PV. The primary flaw with EPV was noticed by 
Lewis Carroll long ago. It is, in fact, the first preference which 
means the most, for many voters, and many voters will only vote that. 
There are two approaches to fix this, and they both involve a runoff 
or further process. Carroll's fix was Asset Voting; the standard one 
is Runoff Voting. The runoff increases voter awareness of the top two 
so they will make a more intelligent choice between them. Static 
preferential analysis simply misses this. Voter opinions change. A 
runoff election is a better-informed election. It also tests sincere 
preference strength, this effect hasn't properly been simulated, I 
expect that TTR's real performance is better than what Smith found 
with it, and that the performance would be even better if the primary 
were an advanced method; indeed Top Two Runoff with a Range primary 
has lower Bayesian Regret than pure Range.

The same could be done with an Approval Primary or a Bucklin Primary. 
I prefer Bucklin, in fact, because it does indeed satisfy the desire 
of voters to start with an exclusive preference for the favorite. 
Bucklin, in my opinion, sufficiently addresses the intuitive need, 
and only begins to negotiate with lower preferences when there is 
majority failure in the first round. Same as IRV, really, except the 
negotiation is more open and therefore less likely to miss a 
compromise candidate, one who would actually please this LNH-desiring 
voter that the one the voter could get without that more open negotiation.

>I do think ordinary electors approach voting for a candidate in a 
>public election differently from how they might approach a
>discussion and deliberative vote in a meeting  -  but "no 
>compromise" can be the order of the day there, too!!

Sure, it can. Not in functional groups, though. We have systems which 
effectively enforce disfunction. I'd say that's a habit we should 
break. Range Voting doesn't eliminate the bullet vote, and it leaves 
it just as powerful, it's a full vote exerted for the favorite. But, 
in fact, we know that many voters will add additional ratings. Some 
won't, either because of ignorance of the candidates -- common and 
normal -- or because of LNH fears, for some, but that, to be strong, 
pretty much has to be associated with strong preference for the 
favorite over all others, in which case the bullet vote is really 
quite sincere.

>   As you suggested in
>your post, it MIGHT be possible to "educate" the electors to see the 
>value of giving effect to compromise and how insistence on LNH
>prevents that.  But my experience leads me to think they would still 
>make their "intuitive" response, based on their attitude to
>their first preference choice.

What you are missing, James, is that voters will do this regardless 
of actual LNH compliance by the system. I think there will be some 
additional vote suppression with Approval, though not much that 
actually affects the outcomes. There will be less with Bucklin, and 
it's entirely possible that Bucklin and Optional PV voting patterns 
will be nearly identical. I.e., lots of truncation, but also 
extensive use of additional preferences in a final election. Not 
necessarily in a party primary, which is more like an initial 
negotiating position. (Extensive? up to about 30%. Actual Bucklin 
experience. Is Optional PV any better than that? If so, how much 
better, and under what conditions?)

> > It is no wonder that a referee, reviewing Woodall's original paper
> > describing and naming Later No Harm, called it "disgusting." (This is
> > reported by Woodall in the paper.) So this is not just my view, James.
>
>The comment by the referee was a personal value judgement.  That 
>comment and that language should have had no place in a professional 
>review of an academic paper.  I am pleased that Woodall published it.

Published the criterion, or published the comment by the referee? 
Woodall was noting that the desirability of the criterion was 
controversial. It's one thing to note a criterion as a 
"characteristic" of a system. There is no value judgement in that. 
But when criteria are considered "desirable," that's a value 
judgement, and there is an implication with many of the criteria that 
they are *desirable* characteristics. LNH and FairVote's invented 
Core Support Criterion are both quite unclear.

The Majority Criterion looks really good, and seems to be an 
expression of Majority Rule, but it actually is not, because 
preference strength has been neglected. Majority Rule is confined to 
a single choice, ideally a Yes/No answer to a question. When there 
are multiple majorities in a multiple-choice election, the majority 
has not actually made a decision between them; tradition is that the 
alternative with the highest vote will win, but, necessarily, this 
allows that the first preference of a majority may not prevail -- 
because it wasn't necessarily expressed by a majority. Runoff Voting. 
Whenever a majority hasn't made a choice, the election fails and 
"must be repeated," in the language of Robert's Rules of Order. That 
means no elimination in pure democratic process; candidate 
elimination is a compromise made in the name of necessary efficiency 
-- it's believed. It's actually possible to fix this, to make 
unlimited balloting practical. But .... there will be opposition, you 
can be sure. There are strong forces opposing democracy, still, 
forces that do not trust the judgement of the people and which wish 
to confine their expression to manageable contexts, where the 
"knowledgeable" or the "more deserving" or just "we" can manipulate 
the larger public.

>I never suggested that this view of LNH was yours alone, Abd.  I am 
>well aware it is shared by quite a number of others, who put
>other criteria above (or well above) LNH.

I essentially discard as not absolute and of lower value all the 
criteria except the one that nobody bothered to name: SU 
maximization. If votes can be taken as expressions of utilities, the 
method satisfies the SU-Max criterion if it always chooses the 
maximizing candidate. Approval satisfies this (with restricted 
votes). Range satisfies it. Bucklin will almost always satisfy it, 
excepting where there is a majority preference that pops up before 
the compromise candidate, the maximizer, has all his or her votes 
revealed. (I.e., this is a majority criterion compromise. Bucklin was 
usually ranked Approval.)

There is an interesting form of Bucklin, I just realized, that was 
actually Range Voting. Oklahoma had a three rank ballot. First rank 
was normal, like all Bucklin methods, if a majority wasn't found in a 
round, then the next round votes were added in. In the Oklahoma form, 
the 2nd rank votes were divided by two, and the third rank votes by 
three. That's a range method, lower preferences are devalued.

I don't think an election was actually held with it, opponents got an 
injunction against it in time, but the reason wasn't the fractional 
votes. It was mandatory ranking. The norm in Australia, and what most 
voting system theorists who've worked with an promoted IRV seem to 
assume will be normal voting. The idea that IRV guarantees a majority 
result is based on this assumption: full ranking. Quite simply, it 
doesn't happen. It doesn't really happen with normal Australian PV; 
the lower ranks for many voters are just noise.

>I am not going to comment of the rest of your interesting post in 
>detail, but I am surprised that anyone should take Bucklin
>seriously.  I, and some of our intuitive electors, would regard it 
>as fundamentally flawed because a candidate with an absolute
>majority of first preferences can be defeated by another candidate.

No, that's not possible with Bucklin. You've misunderstood something. 
You've confused Bucklin with Approval. Bucklin is *ranked* approval, 
in essence, so if there is a majority in the first round, which there 
would be in the situation described, the election terminates. That's 
it. In actual practice, the lower ranked votes were counted, it's 
easy enough to do it, but they were legally moot.

Approval, by the way, guarantees a majority its first preference *if* 
the majority votes it as an exclusive preference. It only allows 
something else if the majority suppresses its preference by adding 
other votes. Bucklin doesn't have that problem. If it's ER-Bucklin, 
they *could* suppress it, but there is little or no strategic 
advantage in doing this unless the preference is very small. In 
actual Bucklin applications, only one vote was allowed in the first 
and second ranks.

Now, I'm glad that Mr. Gilmour has expressed his horror at a result 
contrary to the first preference of a majority. It's quite 
understandable, but I'm surprised that, with all that has been 
written on this list, he still holds this as a position. It's 
preposterous in direct negotiations in healthy society, *and the 
majority will agree." I.e., put the result to a ratification vote, 
the majority will vote to go ahead with what was not the first 
preference of a majority. It's only the process of confining all this 
negotiation to a single ballot that makes it seem weird. We are so 
accustomed to the idea that preference strength *cannot be expressed* 
that we only think about preference as an absolute, with all 
preferences being equal. Above Mr. Gilmour expresses the idea that 
first preference means the most. Sometimes. Not always. What if the 
first preference is weak, the voter has practically no preference? 
"Well, if you ask, other things being equal, and it's fine with the 
rest of us, I prefer A." Or the first preference was forced by the 
method, there was no allowance for equal ranking and the voter had no 
preference at all, couldn't distinguish between the candidates?

FairVote gives the preposterous example of 99% of voters who vote, in 
a Range election, A 100, B 99. And one voter comes along and votes A 
0, B 100, and prevails. Obviously bad result? We don't really know. 
If we want to know, we notice the Condorcet violation and we hold a 
runoff. Then we will know. Otherwise, if we take all the votes as 
sincere -- why not? -- the result of B was actually *slightly* 
better. Very slightly.

A better example would be 51% A 100, B 99, and 49% A 0, B 100. What's 
the best result? Almost certainly, the B result is *much* better than 
A. B deeply satisfies every voter, according to what they expressed. 
Boy, I wish I could get that kind of result (a 99!) more often! But, 
hey, Majority Criterion violation. Do you still believe that A should 
win? The majority -- a slight one -- give up a very small, 
practically indistinguishable preference, so that *everyone* can 
enjoy the result of the election. Anti-Range activists will claim 
that the B voters were insincere. But they clearly had sufficient 
preference to want B to win, more strongly than 100:99! Voters don't 
vote insincerely unless they have some kind of strong preference!

>   Such a result may measure some "compromise view" computed from
>the voters' preferences, but it is not considered acceptable  -  at 
>least, not here for public elections.

Nor here, not yet, nor do I think it should be accepted. But I'd set 
up preference criteria so that when the majority criterion -- or the 
Condorcet criterion -- fail, the public is asked to confirm the 
result. That could be a runoff, or it could be through an Asset 
technique, where the candidates who received first preference votes 
are asked to participate in the runoff as representatives of those 
who voted for them.

James, your "intuitive electors" may have some excuse, and we need to 
address their concerns, but you don't have that excuse. Bucklin was 
extremely popular in the U.S., and the LNH arguments were raised 
against it then, it's not like Woodall invented the concern. I see 
little sign that LNH fears caused the majority failure that was 
moderately common with Bucklin. Rather, advanced voting systems do 
encourage more candidacies, which encourages more vote splitting and 
with normal truncation, majority failure can be common. San Francisco 
was getting, in certain districts, routine majority failure. They 
though they were going to fix this with IRV. In fact, IRV experiences 
the same majority failure, and institutionalizes it and makes it 
appear to be a majority, unless people are paying attention. The 
winner in one district, with over twenty candidates on the ballot 
(single-winner), got less than 40% of the vote. Almost always, IRV 
does not find a majority here in nonpartisan elections where no 
majority was found in the first round. That's the dirty little secret 
for a method which is sold as guaranteeing a majority.

There is no way to guarantee a majority except to coerce the voters.