[EM] Strategic polls in Approval

[Abd ul-Rahman Lomax]

At 08:08 AM 12/30/2006, Juho wrote:
>Here's one simple strategy for Approval that I don't remember having
>seen on this list.

I'm personally focused on practical changes to public election 
systems. For me, the essential question is not whether a system has a 
potential flaw, but whether or not it is better or worse than what we 
presently have. The particular vulnerability alleged here is actually 
applicable to *all* election systems, it happens to be here stated in 
a specific way that could make it appear that AV is more vulnerable 
than, say, IRV or IRRV.

The general vulnerability is that the public can be deceived by false 
information. In this case, it is false information about the 
frontrunners, through a forged poll. Obviously, if the public can be 
deceived about who the frontrunners are, votes may be nudged in a 
system where a common strategy involves determining, first, th 
identity of the frontrunners.

However, elections where this kind of fraud would succeed would be 
rare. Juho proposes:

>three candidates, A, B and C
>- a neutral (and correct) opinion poll tells that A and B are the
>leading candidates and C has somewhat less support
>- some A supporters publish a new forged opinion poll that says that
>A and C are now the leading candidates and B has somewhat less support


This is an election where there is something close to a three-way 
tie. That's rare, for starters.

By the definitions of frontrunner, all three are frontrunners. There 
is risk of the election of any one of them. A prudent voter will consider this.

This fraudulent strategy (which is a strategy on the part of A 
supporters, not of voters), to succeed, must involve a poll which is 
widely publicized and believed. Presumably there will be conflicting 
polls, there is no reason to suppose that A supporters have a 
monopoly on polling. Indeed, we normally see conflicting polls, it is 
routine. But when a poll is drastically different from what the 
majority of polls are showing, we, quite properly, suspect it.

I don't think the poll would have much effect. For this strategy to 
work at all, the election has to be close, there must really be some 
doubt about which of the three candidates will actually turn out to 
be in third place. So voters will consider that.

Essentially, they will rate the candidates (Approval is a Range 
method, really), determine an Approval cutoff, and vote for all 
candidates above that cutoff.

The common voting strategy proposed for Approval is to look at the 
frontrunners, it is assumed that there are two, pick the favorite of 
the two, and then approve that candidate plus all candidates you 
prefer to that candidate, if any. Note that most people using that 
strategy will, quite likely, in such an environment, only vote for 
one, because, by definition, most people prefer a frontrunner over 
all others. *That's why they are called frontrunners!*

Where there are three frontrunners, this strategy does not exactly 
apply. That's the situation proposed. There are really no 
frontrunners in this race.

The protection would be a requirement that the winner obtain majority 
approval. Many elections require this. With that requirement in 
place, a different strategy is proposed that is much simpler and not 
vulnerable to this polling deception, because it does not involve 
determining the frontrunner.

For each candidate, make an independent determination: would you 
prefer the office to remain vacant or for this person to be elected? 
If you'd prefer the office to remain vacant, don't vote for the 
candidate. If you would prefer the candidate to be elected, vote yes. 
The winner, with approval, then, is the candidate most widely 
acceptable, or there is no winner.

I've learned a great deal by thinking of the deliberative process 
used under Roberts Rules as the basic *deliberative* election method: 
sequential nomination and vote. Someone moves that A be elected as 
the officer. If the motion is seconded, it is then open to debate and 
amendment. If I move that B be substituted in the motion for A, and 
my amendment is seconded, *this* can then be debated, and, 
presumably, will ultimately come to a vote. The vote essentially 
determines if the members prefer B to A or not. If they prefer B to 
A, then the amendment will pass. This process continues until all 
reasonable nominations have been made and accepted or rejected. At 
some point, debate is closed, typically by a two-thirds vote, and the 
main motion proceeds to vote. If it passes, we have elected an 
officer. If not, the post remains vacant or is otherwise filled 
temporarily according to the emergency rules that apply when a post 
becomes vacant for some reaso

*This is a Condorcet-compatible method,* but what may be easily 
missed is that preferences, in a deliberative process, are not fixed 
things. This is the difference between deliberative and aggregative 
process. Aggregative process, i.e., voting, simply attempts to take a 
snapshot of preferences, and possibly preference strengths as with, 
for example, Range, at one point in time. It includes preferences 
that would change if exposed to debate. It includes preferences that 
would shift if people knew what others preferred.

Because of our continual exposure to preference systems, and a lack 
of exposure to Approval and Range methods, we tend to think of 
elections as a contest, with voters being opposed to each other if 
they prefer different candidates. But in real organizations, where 
people care about each other and about the health of the 
organization, it can matter very much that a candidate is not merely 
the preference of a majority, but also that the candidate is more 
widely *acceptable* than that. And I have seen voting in small groups 
where the majority set aside its preference in order to make a 
decision that was actually approved by consensus.

Range methods can collapse this process into a single vote, but the 
give-and-take that is typical of deliberative process is missing. 
Range is still an aggregative method and thus is relatively 
inflexible, compared to full deliberative process.

I think we need to understand that we don't use full deliberative 
process, not because the results aren't superior (they would be) but 
because full deliberative process is very time-consuming and, as the 
group size increases, becomes increasingly so. The answer to this, 
aside from using elected representatives, as in the U.S. Electoral 
College as designed, not as actually used, is Delegable Proxy or 
Asset Voting. This allows the group deliberating to be boiled down to 
a manageable size.

I think the Electoral College was conceived to function in this way. 
However, the Constitutional Convention could not agree on how 
electors were to be chosen, so they left it to the states, a loophole 
that political parties soon discovered and used to full effect. 
Leaving it to the states meant leaving it, aside from the states 
amending their constitutions to prevent it, to the legislatures. And 
the legislatures could make decisions about electors by majority 
vote. So the controlling party in each state could simply award all 
the electors of the state to their own party. And they did.

And this has never really been fixed.

Now, as to the polling fraud strategy. If you can fool voters about a 
poll, seriously, not just a few points this way or that, you can fool 
them about other things as well, you can influence them to vote as 
you desire by deceiving them about issues and facts. This is a 
general hazard to which all aggregative methods involving the general 
public are vulnerable.

And poll results are *not* essential for good Approval Voting. I 
stated an alternative strategy that could easily be used in a 
three-way race, and I suggest it is an appropriate one there.

When there are two front-runners, not three, voters can be reasonably 
certain that one of them will win, and, thus, the realpolitik 
strategy of accepting one of them is prudent, even if you don't 
approve of either of them and would prefer to leave the office 
vacant. That is, if you can be reasonably confident that there *will* 
be a winner from among those two, then you will attempt to influence 
that choice. If not, you will not use that strategy. Your favorite 
does have a shot at winning, and so do both of the others. So you 
will vote for your favorite and then whether or not you vote for 
either of the others depends on whether or not you will find your 
favorite among them acceptable if they are elected.

I think this "Yes/No" strategy is an important one and should be 
mentioned in descriptions of Approval Voting. It is actually the 
foundation strategy.... the choice between frontrunners, if neither 
is your favorite, is a detail arising under certain conditions only. 
Zero-knowledge, you would vote the Yes/No strategy, preferring the 
office to remain vacant to electing a No candidate.