[EM] The worst about each system; Approval Preferential

[Abd ul-Rahman Lomax]

At 12:31 AM 5/27/2010, Jameson Quinn wrote:
>As Abd already said, you can avoid the runoff if only one candidate 
>has a majority. Abd's Bucklin proposal tricks many voters into 
>extending more approvals to decrease the chances of a runoff.

Tricks? I don't know if the runoff will cause voters to add more 
approvals or to reduce them. This is what I think: it sets an 
absolute preference based on the perceived utility of completing the 
election, which is relatively small compared to the value of the 
office being filled by the right candidate. It's enough to 
*encourage* a few approvals, but the existence of the runoff may, 
contrariwise, discourage them.

What's wrong with encouraging voters to add more approvals? They 
tried, in Okalahoma to *force* voters to add more approvals. They 
*require* voters in Australia to add lower preferences (in most 
jurisdictions). I don't support that, either one of those. But 
*allowing* it? With, simply, natural consequences either way?

>  My proposal, the one that started this thread, is simpler to 
> describe and count than Abd's, and it makes extending second-rank 
> approval (and thus typically avoiding a runoff) rational for 
> voters. I think that that will be more effective than tricks**.

Bucklin, very similar to what I'm proposing, was widely used for a 
time. We know that some voters don't like being restricted to three 
ranks in RCV. Additional expression, *if voluntary*, is, in my book, 
a good thing. With three ranks, Bucklin starts to get much closer to 
using a Range ballot, and it allows four candidates to be ranked.


>My proposal again:
>
>Voters rank each candidate as preferred, approved, or unapproved.

So you have an explicit disapproved rank? How is this treated 
compared to a blank?

>  If any candidates have a majority ranking them at-least-approved, 
> then the one of those which is most preferred wins outright.

Right. With quite possibly bizarre outcomes. Now, I can see a value 
to it, and that's why, in fact, I want to make sure that there is a 
runoff if the approval winner is beaten by another (by ranking). Why 
I'd want to use first preferences for this determination, only, I 
don't know and don't understand, except that first preference *tends* 
to be stronger preference.

This method is somewhat ameliorated by being ER in all ranks. But 
having three approved ranks instead of two allows far better 
expression of preference strength. It doubles the expressivity. On 
the other hand, as designed, the ballot is balanced. Mr. Quinn, from 
this point of view, incorrectly, perhaps, assigned values to the 
ranks, instead of his 1, 0.75, and 0, it should be 1, 0.5, 0. But 
that isn't used in this present statement of the method. It's simply 
Range analysis.

>  If not, then the two candidates which are most preferred against 
> all others (ie, the two Condorcet winners based on these simple 
> ballots, or the two most-preferred in case of a Condorcet tie) 
> proceed to a runoff

Utility theory would not suggest his pair. Utility theory suggests 
the sum of scores candidates. I only suggest including a Condorcet 
winner because of conflict between utility theory and democratic 
majority theory. If a result is to be based on "greater summed good," 
the majority should accept it.
[...]
I didn't see this note until the end, here:

>**Insofar as voters agree with the statement "I trust society to get 
>the right answer, even if it's not the one I agree with", it's not a 
>trick. Most people don't seem to believe that, though.

It's not a trick in any case. It's quite open and clear. Do you want 
to see a decision made now, or do you prefer it to be deferred? This 
is the choice faced by voters in repeated ballot, it's perfectly 
ordinary. Do they want to complete the election, or do they want to 
keep voting until the cows come home? It creates a certain natural 
force toward compromise, not enough to cause people to abandon what 
is important to them, but to relax their standards *a little.* 
Bucklin naturally does this within a single ballot, so rerunning a 
Bucklin election extends it a bit more, with an opportunity for the 
voter to revise the voting robot instructions that a Bucklin ballot represents.

Basic concept: A Bucklin ballot is a Range ballot where the voter 
places candidates into utility classes; in original Bucklin there 
were three classes all approved, plus a disapproved class. Voter 
placement of candidates in these classes was relatively 
unconstrained, compared to most methods. That is, equal ranking was 
allowed in third rank, and empty ranks were allowed (which is 
significant only for truncation and an empty second rank.) The method 
then simulates three approval elections with declining approval 
cutoff, seeking a majority. I see no reason to prohibit equal ranking 
in first and second rank: we should remember that when we 
unnecessarily prohibit possibly meaningful voter behavior, we cause 
ballots to be spoiled. Equal ranking has an obvious meaning: 
relatively low preference strength, or a strategic decision to equal 
rank because one of the candidates is considered no-hope. This allows 
voters to vote sincerely for no-hope candidates without *any* loss of 
strategic voting power; with equal ranking such equal ranking is 
simply moot for election purposes. But people vote for other purposes 
than electing candidates, witness Ralph Nader voters in 2000.

Bucklin, in fact, provides such good handling of ranks that a voter 
could almost always sincerely rank without loss of strategic voting 
power, so the major effect of allowing overvoting is to make the 
voter's decisions easier. If you have trouble deciding which of two 
candidates you prefer, then equal rank them!

The finer the allowed ratings, the easier the decisions actually get. 
A voter who is obsessed about whether to rate a candidate at 74 or 75 
should get a life. It's already overkill, probably at a choice 
between 7 or 8! It's only a hundredth of a vote in the first case, or 
a tenth of a vote in the second (assuming Range 100 and Range 10)! I 
think that Range 4 (with rating 1 not used) just begins to be decent 
in this respect. Instead of just one or two approved levels, I have 
three, and this is useful even if I only want to approve two 
candidates. I can rate my favorite at the top, assuming it's easy to 
figure out which candidate is the favorite. If I think of two 
candidates as clones, I can top-rate them both (if ER is allowed). Or 
I can minimally approve, rating 2, giving my favorite the best chance 
to win before the votes are collapsed as approvals, or I can rate my 
second best at 3, indicating that there is a preference, all right, 
it's significant, but I will also be quite pleased if this candidate 
is elected. (With rating 2, I'm revealing that I'm actually about 
neutral, neither offended or pleased, the result is roughly what I 
think I have a right to expect.)

I've seen people interested in voting systems assume that a system is 
difficult, because all the strategic implications of each vote, 
because the *exact optimum strategy* isn't necessarily easy to fix. 
But that's based on overthinking it. A good system will amalgamate 
preferences in such a way that when a decision is difficult to make, 
it has little likely impact. Do I rank one candidate above another, 
or equal rank them? If that's a hard decision, either choice is 
probably roughly correct. And in a Bucklin system, it's a small shift 
in the result, not a large one (through Range analysis, in a hybrid 
Range/Bucklin system).

Bucklin/Runoff, with good ballot analysis if a majority is not found, 
frees voters to put together a sincere categorization of candidates; 
in the primary, the basic question is fairly simple. Given what you 
know now, would you prefer to elect this candidate, or would you 
rather wait to make that decision in a runoff? If you prefer to elect 
now, approve the candidate in one of the approved classes. If not, don't.

And then place the candidates in the classes simply: put your 
favorite on top, put the least favorite approved candidate on the 
bottom approved rank. And if there are any left, you consider which 
is better for them: top, middle, bottom approved rating. If that 
decision is difficult, again, you are overthinking. My guess is that 
the difficulty would only involve one rank step. In most Bucklin 
elections, quite likely, all the ranks collapse so which approved 
rank you put the candidate in will only matter if no majority is 
found. And how it matters depends on the exact runoff determination 
method, which is what we've been discussing. If a preference is 
important to you, express it, assuming there are enough ranks 
available. If it's not important, don't express it unless it's easier 
to express it than not. More ranks makes it easier, as long as there 
are enough ranks to rank all candidates, if that's what you want to do.

That is, I'd like to see the ballot be Borda-like, at least as many 
ranks as candidates. But with equal ranking and thus empty ranks 
allowed, which, of course, makes it a Range ballot.

(Which shows, in fact, that Borda is just Range with a restriction on 
the voter, a restriction that causes the basic Borda pathology. 
Restricting voters without good cause is generally a Bad Idea. Tell 
me again why we don't want to *allow* voters to equal rank!)

Voting systems need to be flexible, to handle, sometimes, large 
numbers of candidates. Other times there may be anything from none to 
one, and on up. In San Francisco, I've seen 23 candidates on the 
ballot. Plus some approved write-ins. Three ranks on an RCV ballot in 
that place is very tight, particularly because equal ranking isn't 
allowed. If it were allowed, three ranks might even be enough. But 
once one is thinking this way, IRV is a lousy canvassing method, 
there is no good reason to put up with its flaws.