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

[Abd ul-Rahman Lomax]

At 08:56 PM 12/22/2008, Terry Bouricius wrote:
>Dave,
>
>I think you make a common semantic manipulation about the nature of a
>Condorcet winner (particularly in a "weak" CW example) by using the term
>"wins by a majority."

He wouldn't be the one who invented this practice, Terry.

>  In fact, each of the separate and distinct pairwise
>"majorities" may consist largely of different voters, rather than any
>solid majority.

That's correct, with truncation. This is why it's an important 
reform, one long ago introduced into the U.S. in progressive 
jurisdictions, to require a majority of ballots contain a vote for 
the winner, at least in a primary that can elect if there is a 
majority. The big error that was made, and it was made with Bucklin 
as a runoff substitute -- it was also sold that way, as well as with 
IRV -- was to imagine that, somehow, these preferential voting 
systems were going to manufacture a "majority."

>  This is why I think the Mutual-Majority Criterion is a
>more useful criterion. In a crowded field, a weak CW may be a
>little-considered candidate that every voter ranks next to last.

Sure. But wait a minute! "Every voter ranks next to last." Ain't 
gonna happen unless "every voter" ranks all the candidates. Under 
voluntary ranking systems, that represents "every ballot containing a 
vote for the Condorcet winner." Consider the case that this is RCV, 
three ranks. That's a strong showing! But it *also* isn't going to 
happen in public elections, except possibly with exceeding rarity 
under unusual conditions. The biggest factor in voting patterns in 
public elections is probably name recognition. A Condorcet winner 
with high name recognition is also going to get a substantial number 
of votes. This is not a "little-considered" candidate if everyone 
uses one of their three ranks for the candidate.

Terrill, it's pretty obvious to me that the decision to support IRV 
in the U.S. was made for strategic reasons, not because the system is 
actually superior to other options. IRV is probably inferior 
(different result being a worse result) in probably one out of ten 
nonpartisan elections or so, just compared to top two runoff, not to 
mention better systems (in my view, real runoff voting has a huge 
advantage, such that it's possible that TTR, with write-ins allowed 
-- as is the case in some U.S. jurisdictions -- is better even than 
Range. And what have you and your friends been replacing with IRV? 
Plurality? No. Top Two Runoff, vulnerable because of (1) widespread 
ignorance about the difference between IRV and TTR, and (2) an 
alleged -- and possibly spurious -- cost savings.

FairVote's first "victory" was the San Jose measure that allowed IRV, 
in 1998. The ballot arguments were flat-our wrong. They essentially 
would only be correct with full ranking, which is a Bad Idea in the 
U.S. and is the reason why Oklahoma Bucklin was ruled 
unconstitutional. It wasn't the additive method, it was the mandatory 
full ranking. The ballot analysis by the "impartial" county counsel 
-- who apparently swallowed the propaganda -- and, of course, the Pro 
argument by Steve Chessin et al, very specifically misrepresented the 
majority issue, using "ballots" instead of the somewhat vaguer 
"votes" in the similar San Francisco situation.

IRV *functionally, in nonpartisan elections*, is Plurality. The 
difference must exist, sometimes, when an election is close enough, 
but it is rare enough that we haven't seen it yet in the U.S. in over 
thirty such elections. And since, if it does occur, the vote is 
likely to be quite close, it's quite unclear that IRV would be enough 
better than Plurality *in that context* to make it worthwhile. TTR 
*is* better, clearly, in probably one out of ten elections.

I'm waiting for you to realize just how much of a mistake was 
made.... You and FairVote have been damaging U.S. democracy, 
replacing the only method which is known, in practice, to encourage 
strong multiparty systems, with IRV, which doesn't. That method, Top 
Two Runoff could be made better by using a better preferential voting 
system in the primary, and submitting, to a runoff, ambiguous results 
(such as majority failure, but there are other possible situations, 
such as a multiple majority in an additive system like Bucklin -- 
though, here, there is good precedent for choosing the candidate with 
the most votes). IRV avoids runoffs by discarding the majority 
requirement through a trick definition. Bucklin doesn't follow that 
definition, if there is a majority in Bucklin it is not a trick, all 
the ballots are included and counted. So Bucklin will show "majority 
failure" when IRV can conceal it, to those who don't pay attention, 
by only considering the last-round votes, meaning that many voters, 
who did vote and cast legitimate votes, and who did not necessarily 
truncate, don't count, it is as if they did not vote.

>   The
>phrase "wins by a majority" creates the image in the reader's mind of a
>happy satisfied group of voters (that is more than half of the electors),
>who would feel gratified by this election outcome. In fact, in a weak CW
>situation, every single voter could feel the outcome was horrible if the
>CW is declared elected.

If so, they were totally foolish to rank that candidate. Ranking a 
candidate should be reserved for acceptable candidates, mediocre *at 
the worst*. Not horrible. And this is what voters actually do with 
Optional Preferential Voting. Only at most something like a third of 
the voters add additional preferences, they only add them when they 
think them worthy of being elected.

Terrill, you are making up arguments to defend the indefensible: 
"Condorcet winner" means that every other candidate would lose in a 
pairwise election with that winner. (If not, voters are using odd 
strategy, or perhaps they are forced to rank all candidates, as in 
Australia, so equal-bottom is not available.) Borda detects the 
problem, giving the next-to-last candidate only a very weak vote, 
and, indeed, this is one reason why Borda fails the Condorcet 
Criterion.... The problem with Borda is that it doesn't allow 
differences in preference strength to be shown, using the candidate 
sequence as a very rough approximation of preference strength. But 
that's fixed with Range, which is otherwise, really, the same as Borda.

Now, using hypothetical scenarios that would be impossible in real 
elections under real conditions is a means that can be used to 
confuse and deceive the unsophisticated. Ask an expert if the system 
is vulnerable to this, the answer will be Yes. Only if the expert 
answers with more than what the question addressed will the inquirer 
learn the truth. "Yes, but this scenario is, in practice, impossible."

Yet when much more common situations, such as Center Squeeze, are 
raised with IRV, the answer is very different. "It's rare, prove that 
it happens!" Center Squeeze, though, is a very well known problem 
with sequential elimination, it's remarkable that Robert's Rules of 
Order chooses to mention it as a problem. The problem happens with 
reasonable frequency with Top Two Runoff, and the principles are the 
same. *In this way,* IRV simulates TTR, though, in fact, it is a 
little better in choosing among the remaining two. IRV would not have 
elected Le Pen. But it missed the very clear, and broadly supported, 
Jospin, who would have won against Chirac, and not by a small margin. 
If Jospin had gotten 1% more of the vote, the runoff would have been 
between Chirac and Jospin, the voter turnout would have been lower 
because of lower preference strength, but I'd have predicted, from 
what's known, 70% or more of the vote for Jospin.

These are real situations, not theoretical scenarios where voters 
vote some totally stupid strategy.

>  Using a phrase like "wins by a majority" creates
>the false impression that a majority of voters favor this candidate OVER
>THE FIELD of other candidates AS A WHOLE, whereas NO SUCH MAJORITY
>necessarily exist for there to be a Condorcet winner.

That's correct. And that is why one submits such elections to a real 
runoff of some kind, which is the only way to get a majority. It 
almost always does. It only fails when there are write-in candidacies 
allowed. It's possible by using better methods for the primary and 
runoff, to allow write-ins without harm, where a real majority would 
be found in the vast majority of elections. Want to make that a 
guarantee, it must be an unlimited series of elections, no 
eliminations. And the only way I know of to do that is to elect 
deliberatively, and only Asset Voting can pull this off on a large scale.

But Top Two Runoff approaches it, except for Center Squeeze, which 
can be fixed by using a better method for the primary. How about a 
runoff between a Condorcet winner and a first preference winner? But 
there should be further process whenever acceptance of the leader is 
not shown on a majority of ballots. That's Robert's Rules, you know, 
even though you argued derisively against it for a time.

Limiting it to the top two is a compromise, a poor one with a 
plurality primary, in some cases. Better methods could make a better 
choice of top two. I've suggested that, in Range, when the Range 
winner isn't approved by a majority (as explicitly defined on the 
ballots), or when there is a candidate who beats the Range winner 
pairwise, using preference analysis, there be a runoff. This, from 
strong theoretical grounds including simulations, will improve 
results over Range, which otherwise is the leader with single 
ballots. That specific combination (range/pairwise) hasn't been 
simulated, but top two Range has.

>  The concept of
>Condorcet constructs many distinct majorities, who may be at odds, and
>none of which actually need to like this Condorcet winner.

Sure. If there is full ranking. But, you know full well, truncation 
is the norm. RCV doesn't even allow more than three ranks to be 
expressed, and still there is a lot of truncation. I really should 
look at those San Francisco ballot images. What I want to see is how 
many voters don't add any ranked votes to votes for the frontrunners. 
My guess is that *lots* of them don't. If they do, those are 
candidates that the voter actually wants to support in some way, not 
realizing, perhaps, that the vote will never be counted.

Isn't this a bit disturbing? Quite likely, most of the votes cast in 
an IRV election aren't ever counted. That's not true with American 
Preferential Voting, Bucklin.

Watch for arguments against IRV to start using the fact that Bucklin 
was invented here in America and was, indeed, known as the American System.

>  I am not
>arguing that the concept of "Condorcet winner" is not a legitimate
>criterion, just that its normative value is artificially heightened by
>saying the candidate "wins by a majority" when no such actual solid
>majority needs to exist.

That's correct, by the way. However, if there is full ranking, and 
based on the preferences and votes as expressed on the ballot, it 
does indicate that the winner would, indeed, win by a majority 
against any other single candidate.

It's true, though, that if there were an actual TT runoff, the voters 
would give sharper consideration to the two candidates on the runoff 
ballot. Voters change their minds. Some of the bottom-rank Condorcet 
votes may have been strategic, faced with the reality, the voters 
might reconsider.

I assume one thing: voters do, almost entirely, vote sincerely in 
IRV. It appears that, historically, they also did this with Bucklin. 
The effect of preference strength on voting patterns has largely been 
neglected, instead, a concern for "Later No Harm" has been expressed, 
as if this were some universal voter concern. In fact, it depends on 
preference strength. If the voter has little preference strength for 
the Favorite/2nd preference pair, the voter is quite likely to add a 
ranked vote. If it is strong, not, unless the voter is clear that the 
Favorite hasn't a prayer, and even some of these will still truncate. 
That's what they do with Plurality (vote contrary to interest in 
outcome), that's what they do with IRV, when they can, and they 
likewise did it with Bucklin.

Voters with strong preference, and particularly if their favorite is 
a frontrunner, are unlikely to add additional preferences unless you 
coerce them, and coerced votes are the kind that could show the weak 
preference being described for a Condorcet winner. Solution: don't 
coerce voters!