[EM] reply to venzke - range "random skewing" effect is not a problem

Warren Smith wds at math.temple.edu
Mon Mar 12 18:51:53 PDT 2007

>Ritchie: Warren, you're assuming that the same number of voters will act
strategically under different methods again.

--I simply presented data.  You may have made such an assumption.  Not I.
I did not, anywhere on the web page  http://rangevoting.org/StratHonMix.html
state  "we shall make the assumption the same number of voters will act
strategically under different methods."

But anyway, relevant to this question:
(a) where if anywhere (besides your intuition alone) is evidence that Condorcet voters
are more honest than range voters?
(b) let's COMPARE range (50-50 mix of honest & strategic voters) versus Condorcet
with 100% honest voters.
I just did an IEVS run to do that (by the way, you too can do these runs - IEVS
source code is public)  using the same parameters as the green table on the web page
What happens is the Condorcet methods yield regrets between
0.1127 and 0.1331 (for Black's Condorcet method and UncAAO respectively, which are the
two best Condorcet methods in this run) and 0.1783 and 0.1821 (for SmithIRV and Raynaud
repsectively, which are the two worst Condorcet methods in this run).
(Error bars are +-0.005 or less.)
Meanwhile range with 50% honest and 50% strategic voters yielded regret
0.1633, while range+top2runoff yielded 0.1479.

So: even under circumstances devised to MAXIMIZE Venzke's
"random skewing due to some exaggerating, some not" effect, range does
comparably to Condorcet methods when the latter alone enjoy 100% voter honesty.
I suggest to you that actually, Condorcet voters in the real world are unlikely to
be 100% honest.  But even if they were, range doesn't look bad.

OK?  So fine.  Make an assumption Condorcet voters are all little paragons
of honesty.  Little angels.  Meanwhile the evil range voters are 50% nasty strategic
exaggerators out to screw over the poor little innocent honest ones (other 50%).
Fine.  Range still does comparably to Condorcet methods, say 29999 simmed
elections, in circumstances designed specifically to make range voting look bad.

Any questions?
(I will soon try to edit the web page to incorporate the data from this new IEVS run.)

>Venzke: These simulations purport to show that Range does relatively well by SU
when voters are a mixture of strategic and sincere. This is pretty
tangential to what I wrote.

--what Venzke wrote was:
> >Venzke: If I don't want to assume that voters will courteously
> vote sincerely (even when this limits their power to affect the results),
> then I wouldn't use Range, as the result will be rather randomly skewed
> based on who chose to exaggerate and who didn't.

--My simulations addressed exactly this. They were not "tangential."   They were
"a study designed to examine exactly this question." The plan was to set up a situation
with maximal "random skewing" due to some voters (50%) "exaggerating" and some (50%) "not,"
and choosing who was who "randomly" by coin toss, and having a small total # of voters
(61 and 13 voters in the two tables) exactly to make sure there was a large typical
variation in the numbers of honest & strategic voters in each political camp.
EXACTLY the situation Venzke was worried about.

Warren D Smith

More information about the Election-Methods mailing list