<div class="gmail_quote"><br><div class="gmail_quote"><div class="im">2009/8/30 Warren Smith <span dir="ltr"><<a href="mailto:warren.wds@gmail.com" target="_blank">warren.wds@gmail.com</a>></span><br></div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<div>> "utterly false" is a bit strong language<br>
<br>
</div><div class="im">--No: "absolutely" "unbeatable" is what is "strong language."<br>
And it is false. And what really pisses me off, is you accusing me of<br>
being false, when it was you all along being false, and indeed I had<br>
publicized the exact opposite of what you falsely said and falsely<br>
attributed to me. The correct response is<br>
to apologize (jerk), not to come back accusing me of "strong<br>
language"! Holy cow!</div></blockquote><div><br>I'm really sorry if I was being unintentionally abrasive, to the point which would have justified you calling me names. I can honestly say that I have a lot of respect for you and I hope we can keep any further debate in the realm of ideas and not name-calling.<br>
<br>I am not accusing you of being false on this point. I am saying that you are using strong language, extending a good result on RNEM to be generally applicable when it's not. <br><br>Your BRBH system is derived in the case of 3-candidate RNEM. Say I create another elections model, in which voters have a bimodal utility distribution for the 3 candidates, with a gap in the middle. Then, a (-1, 0, 1) vote is Bayesian evidence that the voter is one of the 1/4 who is sampling all three candidates from the same utility mode, and therefore evidence that the BRBH-like "denormalization" weighting factor should be at a minimum, not a maximum. A vote that's more like (-1, 0.9, 1) or (-1, -0.9, 1)is, on the other hand, evidence for a high "denormalization" factor. So in this voter model, BRBH is significantly worse than plain Range.<br>
<br>Yes, it is probably false that Range is "absolutely unbeatable" in BR with 100% honest voters. But without specifying the model, it's going a little too far to say that it's "utterly false". Anyway, I'm sorry we are getting distracted here. The original point I intended to concur with (and <b>I'm sorry</b> I didn't clarify in my concurrence) was that it is not really worth trying to beat Range's BR with honest voters, in simply practical (not mathematical) terms. Can we agree on that?<br>
</div><div class="im"><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"><br>
<div><br>
> I completely reject your model of strategic voters. It is simply not true that, if you surveyed all voters their honest opinions of who the two "frontrunners" are, all of them would agree in 100% of elections.<br>
<br>
</div>--Ok, the truth is not 100%, it is "over 98%."<br>
Indeed, it is possible (proven by test) to predict all important US<br>
races with >98%<br>
accuracy over 1 year ahead of time. Proof: see remarks on Ron Facheux here:<br>
<a href="http://www.rangevoting.org/NonVoters.html" target="_blank">http://www.rangevoting.org/NonVoters.html</a><br>
Also: note the percent of third-party winners in the USA in such races<br>
is below 1%.<br>
<br>
OK. Now, is this difference between 100 and 99% really a good basis<br>
for your "complete rejection"? Golly. Gee willickers. Total<br>
unrealism! You ARE finicky.</blockquote></div><div><br>Your own graphs show that, for infinite numbers of voters, any epsilon of strategic voters causes a discontinuous increase in BR. If you graphed the "voters can predict frontrunners with x% accuracy" versus number of strategic voters, it would have a similar discontinuity. With finite numbers of voters, it would just be nearly discontinous, rising to 100% as the strategic voters become something like the square root of total voters. If you found 99% on that graph, and then projected down with that same fraction of strategic voters to the similar graph for some non-plurality election system, I think you'd get a number far less than 99%.<br>
<br>In other words, I think the empirical 99% is nearly entirely explained by the special flaws in plurality, and that, given those flaws, the fact that it isn't 100% is, if anything, evidence that a two-frontrunner strategy model will not work for non-plurality systems.<br>
</div><div class="im"><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"><br>
<div><br>
>I propose (and you have not responded) a different model of strategy.<br>
1) Choose an underlying voter utility model which behaves roughly like<br>
reality. This is an area for research, and my own incomplete ideas<br>
</div>don't fit here...<br>
<br>
--actually, I already did this, and did it before you came along and<br>
said this.<br>
(See a common theme? 'Cause I'm noticing one.)<br>
IEVS already includes "reality-based utilities" based on a dataset of<br>
real world elections.<br>
It turns out, when you run the experiment, that using reality-based<br>
utilities leads to little-to-no discernible difference in results,<br>
versus using fantasy-based utilities like<br>
I had before.<br>
<div></div></blockquote></div><div><br>Sorry, I didn't acknowledge that you've done the best work I know of in this realm. I personally think we all owe you a debt in this regard: thank you. Still, this is only step 1 of my proposed scheme.<br>
</div><div class="im"><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<div><br>
>I can make Range behave like "random candidate" by assuming that candidate X's voters strategically bullet-vote and all other voters vote honestly. (For infinite numbers of voters in RNEM, you just assign a random epsilon(k) fraction of bullet voters to each kandidate, and then whichever one has the highest epsilon(k) wins.) Both assumptions are unrealistic worst-cases but arguably a significant drain factor on the systems' overall quality.<br>
<br>
</div>--this seems unlikely to be a realistic model. But if you think it is<br>
realistic, then fine.<br>
Put it in IEVS, and compare range to other voting systems under that model.<br>
In particular, under the very model you just proposed, you will find<br>
that Condorcet<br>
systems also elect the highest-k guy. Congratulations. You just<br>
proposed a new model and found range and Condorcet and random<br>
candidate all behaved the same under that model. Which is fine, but<br>
seems little basis for "[range] is not as great as you depict it."<br>
I mean (since you seem to be a Condorcet supporter?) wouldn't it be a basis<br>
for saying Condorcet is not as great as YOU depict it?<br>
<div><div></div><div></div></div></blockquote></div><div><br>OK, a couple of points:<br><br>1. I did ask you for a link to the source code of IEVS, please. I'd like to do this.<br><br>2. If you follow the strategy model I gave, it is not obvious that Condorcet models do this. Even if different ideological groups of voters have different strategy thresholds, methods which are more resistant to strategy can either discourage strategy altogether or prevent it from having such an absolutely decisive impact that the most-strategic group just wins, period.<br>
<br></div></div>Jameson<br>
</div><br>