[EM] Why the concept of "sincere" votes in Range is flawed.

[Abd ul-Rahman Lomax]

At 03:34 AM 12/1/2008, Kristofer Munsterhjelm wrote:
>Range and Approval might not be insincere (if we accept your 
>definition), but they still require voters to use strategy - that 
>is, to keep the votes of others in mind when they're voting. In 
>Approval in particular, this is very important (consider the 
>Bush-Gore-Nader situation - do you vote for Nader, or {Nader, 
>Bush}?). Therefore the method will work to the degree that the 
>voters know this information (from polls, etc).

Actually, remember, Range works *best* when voters vote using 
zero-information utilities. That is, the method works better, selects 
the candidate with the best overall satisfaction, if voters don't 
know how the other voters will vote, but only know the candidates.

Further, while a voter may improve their personal expectation by 
voting with maximum strength in the "important" election, i.e., for 
one of the frontrunners or *maybe* two if there are three 
frontrunners, the improvement is actually small, relatively speaking.

If a voter has sincere relative utilities for three candidates as 0, 
1, 2, then the voter's "fully sincere" vote would be just that in 
Range 2. What's the maximally effective vote? Turns out, 
zero-knowledge, it's the same, but voting Approval-style has the same 
expected utility. There is another difference. Voting Approval-style 
has greater variability. It's more likely to elect the favorite than 
the sincere vote. But it also is more likely to elect the worst 
candidate. These balance each other out.

But what some seem to be concerned about is that a voter, given 
knowledge of how others will vote, can enhance their expected result. 
I'm not sure there is any voting system worth using where this is not 
true. In the limit, you know all the other votes. Now how do you 
vote. If you can change the result, the electorate has essentially 
said, "We don't know what we want, we are split between this and 
that. So you decide." Should the voter ignore that?

In real elections where the voter's vote makes a difference, almost 
always, the voter can push it to *this* winner or *that winner.* With 
Range, a voter can actually flip a result, but with Plurality or 
ranked methods, the most the voter can do is turn a tie into a win or 
a loss into a tie.

The real situation is, of course, in between the zero-knowledge and 
full-knowledge situations. Polls, though, are rarely far enough off 
that the a voter who has been paying attention will be blindsided, 
and, remember, with Range, the most the voter can do is exert 
Approval-style voting power, and the Approval votes are correct, 
i.e., they do express real preferences, not fake ones. They simply do 
not express all the preferences.

The shift away from optimal cannot be, practically by definition, a 
major one. And, remember, we can test Range results by incorporating 
an explicit "approval" cutoff. I've suggested that any rating higher 
than 50% would be considered "approval," i.e., the voter is willing 
to accept that result. So we can require, as do many elections, a 
majority to win. The various pathologies that *might* rarely affect 
Range (mostly due to strategic voting that isn't sound but is 
deceived -- avoid this by voting sincerely!) are likely to be 
detected by this and can then be fixed in a runoff.

>If that's true, why not just lighten the load on the voters? Why 
>should the voters have to know whether polls are accurate, who's in 
>the lead, and so on? Use a computer to strategize instead. DSV.

Because, remember, we get better results if voters *don't* vote 
"strategically." What it seems you want to do is to institutionalize 
a reduced function, in the name of being "fair."

That's a bad way to be fair, pull everyone down, by distorting *all* 
the information, instead of some of it.

Voters need to know this: don't "approve" a candidate if you would be 
displeased if this candidate won! This is one reason why having a 
majority requirement is important. Voters can back off a bit, have a 
little less need to vote for other than the favorite.

With Bucklin, of course, they can vote for their favorite, add other 
approvals as well which are distinguished from the favorite, or, for 
that matter, not, if they are seriously exercised about Later No 
Harm. (I'm not, I want the best winner, overall, not my personal 
favorite, I don't have any opinion that I'm necessarily a better 
judge than the rest of the voters. But I also think I should participate!)