[EM] encourage dishonesty / range / WDS reply to robla

Warren Smith wds at math.temple.edu
Thu Aug 11 18:54:50 PDT 2005



>Robla:  Rob Lanphier wrote:
>> Hi Warren,
>> I'm interested in Range Voting, since it appears to be popular among
>> many electoral reform advocates here.
>>
> I too get the impression that it is generally considered among the  
> best voting method possible, BUT for one truly fatal flaw. It  
> suffers from the fact that it encourages dishonest voting. Why? It  
> is one of the voting methods in which voters are able to  
> effectively vote strategically with zero information about how the  
> other voters are going to vote - basically  they merely provide the  
> highest possible ranking to those candidates they want to see win  
> and the lowest possible ranking to those they don't. 

REPLY BY WDS:
--That is an interesting but in fact false point.
Specifically, it is NOT possible to determine your strategically-optimal range vote
working without any information about what the other voters are doing.

--Nevertheless, although technically false, your point does carry a certain amount of 
punch, since it feels possible to produce pretty good though non-optimally good
strategic range votes often, without
knowing much (or anything) about what the other voters are doing.

--If you want to compare with IRV, condorcet methods, etc, then you have to ask,
how often it is possible with them to  to produce pretty good though non-optimally good
strategic range votes often, without
knowing much or anything about what the other voters are doing?

But the trouble is, this question is kind of ill-defined and hence not really answerable.

And in reality voters WOULD know something about what the others were doing, too.
Further muddying the picture.

I think what matters more in practice, and what is (closer to being) answerable 
(i.e. by doing experiments on humans) is:  "what are humans actually going to do?"
I believe that in actual practice, Joe Public is often going to exaggerate
in his vote to cause the two frontrunners to be the max and min possible.
This in fact is the best strategy generically for a large class of methods, and is sometimes
but not always the best strategy under IRV and condorcet.   I claim
that a significant fraction of Joe Public is not going to give a flying damn
what you tell him best strategy is or isn't, and is not going even to pay attention
to how the method works (especially if that method is e.g. Schulze beatpaths).
Joe Public is just going to do whatever he damn well pleases, which is 
in some significant fractions each: (1) To be honest, (2) To max and min whomever
he believes are the 2 frontrunners, (3) other ideas (4) mixture of the preceding ideas.
If (1) then range is best by test.  If (2) then range is best by test.
If (4) means mixing 1 and 2 then presumably range is stil best, tho really that needs to be
studied (in the to-do list for when & if I redo my big sim study...).
If (3) then unclear.

I did do a study on real humans.  It did appear they were doing 1, 2, and 4 where
4 means a mixture of 1 and 2.  A very interesting finding of that study was that
real humans although they are tremendously strategic if asked to plurality-vote
(>90% apparently) are tremendously nonstrtageic is asked to range-vote (<24% apparently).

This is a psychological fact.  This fact makes range voting a lot better system
than it would appear to be based purely on mathematics.
(And it already appeared quite good.)

To a certain extent RL's whole argument at the start of this email could
be said to be "I prefer methods that are hard to understand because then
others will not undestand them either and everybody will just give up
on trying to find the best strategy and hence will (as a last resort)
just be honest."    If you think about that argument, I do not think you will
like it.  Also, voters do not need to find the best strat, all they need
is the find a strat they think likely to be as good or better than honesty.
I think in IRV that is exactly what they do do, and that is exactly why every IRV
country is a 2-party dominated country.  I think it likely that every condorcet
country will also be a  2-party dominated country for the exact same reason.

wds



More information about the Election-Methods mailing list