[EM] Strategic polls in Approval

[Abd ul-Rahman Lomax]

At 04:19 PM 1/7/2007, Juho wrote:
>It is also possible that the three leading candidates would come from
>the same party. [etc.]

I find it odd that IRV proponents would claim that IRV can save money 
by eliminating the need for primaries, and eliminating primaries 
could make more likely the conditions under which IRV would select a 
winner against the preference of a majority.

As to false polls misleading voters, if an election is close enough 
between three "frontrunners" I don't think it is at all obvious how 
voters will behave. The false poll could backfire, motivating more 
voters to bullet vote. After all, that is one of the rational 
strategies, and it directly contradicts the vote suggested by what 
was claimed here to be standard Approval strategy.

I would prefer to think of Approval Voting *method* and then strategy 
as a twist on this.

The basic method is to select and vote for all candidates who one 
judges as acceptable as winners. This is why it is named "Approval." 
If voters are fully honest in this, and make reasonable accomodation 
for the views and satisfaction of others (In other words, A is my 
favorite, but I know that B is preferred by many who don't like A so 
much, and I think of B as a reasonable choice for the office as 
well), then the method clearly works to find a good winner.

One way to describe the process simplifies it by placing candidates 
on a linear spectrum and setting an Approval cutoff. In Range, this 
has a specific meaning.... I'll call the Approval cutoff the AC.

However, the reform would be coming into a contentious arena, where 
people are accustomed to fighting for their favorite, even if the 
preference is actually small. They aren't going to move to Canada if 
B wins. They aren't going to need medication for depression, at least 
not for that reason....

So then comes strategy. Who are the likely frontrunners? The voter 
needs to know this because not voting for a frontrunner is likely to 
be a wasted vote. By definition. If the frontrunners are clear, and 
there are two of them, the strategy is obvious, and is what has been 
recommended. If there are three, i.e., the election is reasonably 
close to a three-way tie, in expectation, then one has a choice:

Set the AC such that one of the three is approved, then add any 
candidates preferred to that one.
or
Set the AC such that two of the three are approved, then add any 
candidates preferred to that one.

For a poll to move a candidate from not-close to frontrunner status, 
it would have to be drastically distorted. As I've mentioned, there 
are more subtle, more serious, and less provable forms of lying than 
this. I really don't think it's a matter for special concern.

As I mentioned, it could backfire. We don't know whether distorted 
polls like this would improve or hurt a candidate's chance of 
success, because only one of the possible effects was considered, one 
which is thought to move the result in a direction favoring those who 
distorted the polling. We would also have to consider the effects 
which could move the vote in the opposite direction.

Polls would not seriously affect the internal Range rating which 
underlies Approval strategy. My impression of the desirability of a 
candidate winning does not depend sensitively on what others think. I 
want a candidate to be broadly acceptable, but the difference 
between, say 40% acceptable and 60% acceptable would not have a large 
effect on my opinion of the candidate. That is, very low 
acceptability *would* lower my rating, and very high acceptability 
would raise it, but the midrange is of little effect. In the polling 
fraud scenario, the fraud really only nudges the relative 
relationships of three candidates who are probably about equally 
acceptable. I really doubt that it would cause me to shift my vote.