[EM] Three Stage Approval Election

[Abd ul-Rahman Lomax]

At 02:37 AM 6/7/2006, Dave Ketchum wrote:
>Is this trip necessary?

Yes. At least as necessary as any discussion of ideal election 
methods, actual election methods, and possible intermediary steps.

>I claim not, for it is not up to competing with Condorcet - or even IRV,
>which usually gets the right victor.

First of all, "usually" means that there are exceptions. Those 
exceptions can be devastating. IRV, in particular, could fail to 
elect the best compromise candidate, and when compromise doesn't take 
place in an election process, it can take years of civil war to bring 
people to the table and to peace.

Secondly, Condorcet and IRV are far more complex than approval. 
Approval is probably the absolute simplest election method after 
plurality. It is clearly an improvement over plurality. It takes only 
a rule change to allow it; it actually would be allowed and 
implemented if there were not rules discarding overvoted ballots, 
rules that are clearly undemocratic.

We should all be behind Approval, I'd suggest, even if it is not 
perfect. It is a great step in the right direction, greatly reducing 
the spoiler effect.

While the staged election process proposed in this thread is 
certainly not as simple as basic Approval, and therefore probably 
less practical, it is, in my view, a brilliant solution to one of the 
basic problems with Approval: good Approval strategy relies on a 
reasonably good judgement on the part of voters as to the likely 
winners, if the voter does not vote. An Approval voter, *just like a 
plurality voter under present conditions*, needs to know who the 
frontrunners are, to minimize the chance that the voter's vote will be wasted.

Now, I think there are better solutions, myself, Asset Voting, and, 
in particular, FAAV, my simplified version of Asset Voting, being one 
of the best. But that does not make the discussion of lesser methods 
useless. Among other things, it brings out the issues underlying 
disagreements about election methods. We have discussed this before, 
and have heard the argument presented by Mr. Ketchum before, but, 
obviously, we need to continue that discussion.

>Approval still has a basic weakness.  Easy enough for Approval to be told
>acceptable vs unacceptable, but Approval has no way, even in this
>variation, for me to say acceptable, unacceptable, and between - those I
>would settle for IF AND ONLY IF all the candidates I consider acceptable lose.

That is correct. Approval forces the voter to make a black and white 
decision: Approve or Not.

However, this is what I think Mr. Ketchum misses. The underlying 
problem here is Single Winner. For a Single Winner election to 
represent a desirable outcome, I suggest, the compromise involved in 
selecting that single winner, given all the various preferences of 
the electorate, must be approved, to some degree, by the electorate. 
Or else what we really have is minority rule, which, historically, 
has been a major disaster. The spoiler effect is a very serious one, 
and, among others, can be credited with the election of Adolf Hitler, etc.

Condorcet methods ostensibly solve this problem, but there is also a 
limitation to Condorcet methods. While the voter ranks candidates in 
a Condorcet compliant method, the gap between ranks is not expressed. 
As a result, a minor but common gap can cause a true compromise 
winner to be passed over in favor of one who is merely the favorite 
of a sufficiently large faction. Condorcet methods do not require 
voters to make the judgement of "How acceptable is this compromise?" 
Approval does.

Approval allows voters to make a compromise judgement. With a more 
sophisticated method, that judgement could be more refined. Approval 
is a Range method, with restricted range, essentially binary. 
Increasing the Range increases the level of judgement possible. The 
method which I called A+/PW (Approval Plus, counted Pairwise) is a 
Condorcet-compliant method which adds a single rank, Preferred, in 
addition to the normal Approval ranks of Approved and Not Approved. 
(Basic Approval Plus has this rank, but does not use it to determine 
the winner, only for statistical and campaign finance and similar 
considerations.)

*Any election must find a compromise winner. To be an ideal 
single-winner method, it must find the ideal compromise. What does 
"ideal compromise" mean? Condorcet methods with more than two or 
three ranks don't really address this question, unless they have a 
method of expressing what could be called "ranking distance." I think 
some Condorcet methods attempt to infer this from vote patterns.

But with Approval, the judgement of how to compromise is made by the 
voters, and the election outcome hangs on how the voters make that 
determination, collectively.

So, yes, this is a limitation of Approval, but it is also a strength, 
potentially. It depends on the relative harm and benefit of the two effects.

>BUT, assuming Approval is the best we can do:
>
>"randomly chosen" is tricky - even done truly, voters will wonder if
>selection was biased for a purpose.

Yes, it's tricky. But we don't have to solve that problem here. If 
the method is a good one, presuming that this problem could be 
solved, then that particular problem could be addressed. Since I 
think it is soluble, I wouldn't make this the reason to avoid the method.

You know, there has been a lottery system in the U.S. at various 
times for the draft. I know a lot of people who hated or resisted the 
draft, but I don't recall any allegations that the lottery was 
specifically biased, except in open ways: that is, you could get 
deferments for various reasons, or could otherwise avoid the draft. 
Remember a certain son of a prominent politician who was able to get 
a fairly cushy assignment to the Air National Guard? And then to get 
leave from that to help a certain congressman in his election 
campaign, etc., etc. This was not a problem with the lottery.

Lotteries could be conducted quite openly, and locally. A locality 
would have a quota, a number of voters for the first ballot, and 
those voters might be chosen literally out of a box, by a person 
broadly trusted in the community, with the box contents verifiable 
before and after the drawing.

It could be done. That's the point.

With any election method, we can find some theoretical objection. If 
we don't like the method for other reasons, we may describe this 
objection as if it were an insoluble problem. It's a red herring. If 
the staged-election method is a good method, the problem of random 
selection of voters is soluble.

And it would make a good circus. That is, it would raise public 
interest, I expect, in the election process. It would be exciting, 
like a race. Frankly, it might be quite interesting to have more 
stages. "Smith was in the lead by 2%, but Jones has nosed past him by 
0.1%. This is a real race, folks!"

In a close race, I expect that total election turnout would be 
greater. Most people think that's good. Absent something like 
Delegable Proxy, I agree.

>Voters will puzzle over whether the complications make sense.

They will puzzle over any election method change.

>Messes up campaigning, for candidates need to try to have today's voters
>aware of them today.

A plus for the method, in my opinion. It may even be desirable for 
the sample voters to be in a publicly known set. As a random sample, 
the content of the campaign material wouldn't be different, but 
mailings just to those voters would be less expensive, so that kind 
of campaigning would get less expensive. The media would rebroadcast 
that as news without requiring candidates to pay for advertising. I 
don't have a crystal ball, but, my point is, this could be an improvement.

As with many of these issues, we won't really know for sure until it 
is tried. This method could be tried in a fairly small jurisdiction. 
It might take legislation to enable that, but I don't see any 
constitutional issues, except perhaps with Presidential elections, 
I'm not sure about that.

>Voters will have problems remembering when to vote.

Scraping the bottom of the barrel, there. A voter who has trouble 
remembering when to vote, having received a voting card or ticket, 
this is the person I least want to be voting. It would be all over the news....

The first poll might be quite a bit less than 10%. 1% might be quite 
enough, if the sample is truly random. (In this case, more stages 
would be used, almost certainly.)

>On a normal election day multiple races are attended to at each precinct,
>with a dozen races in perhaps 4 different districts - how does this scheme
>fit in?

Well, they would be included in the stages. I see no reason why not. 
That is, the ballot would stay the same for all stages. There is no 
extra cost for this, as far as I can anticipate. Because of the 
possibility that a margin would be great enough that statistically it 
is nearly certain -- or even impossible -- for the outcome to change 
through the last stage, indeed, costs might be lowered by eliminating 
that stage and replacing it with a mail-in ballot that would be used 
for other purposes, such as campaign finance. But this wouldn't 
happen, I think, with many-candidate ballots. Still, it could save 
voter time, perhaps.

But, to my mind, the value of a good election outcome is much greater 
than the costs for any of the election methods I've seen. Cost 
arguments are, again, a red herring. Election costs are trivial 
compared to the sums that will be handled and allocated by the 
victors; the most significant election cost, which is often 
neglected, is voter time at the voting booth and waiting to get 
there. This cost is enormous, it dwarfs, in fact, in value, what is 
normally spent on campaigning.