<HTML><BODY>
<div> From: wds@math.temple.edu<br>

><br>

> Sorry, I appear to have been an idiot.  Peter de Blanc<br>

> answered my complaints at<br>

> http://www.spaceandgames.com/?p=8<br>

><br>

> and it looks to me like I NOW have to agree with Forest Simmons that<br>

> this IS a great new contribution to voting theory.<br>

> Also, I showed there in my comment how to generalize their scheme<br>

> by adding a parameter P.  It looks <span class="correction" id="">liek</span> the best P is P=0.99 or<br>

> so, not P=0.5 (their value) or P=0+  (my value from before).<br>

                                                                                                                                                           
 <br>

I wonder if it might be worth having the voter write their honest<br>

utilities on the ballot.  The '<span class="correction" id="">backend</span>' would then convert that<br>

vote into an <span class="correction" id="">optimised</span> ballot for inclusion.<br>

                                                                                                                                                           
 <br>

If it <span class="correction" id="">truely</span> is the optimal strategy to be honest, then there would<br>

be no incentive to lie.  However, it would mean that very low<br>

probability of winning candidates would not get lost in the noise.<br>

                                                                                                                                                           
 <br>

For example, they could just have people write their truthful utilities.<br>

The <span class="correction" id="">backend</span> would then just get the square-root of all their utilities<br>

and then <span class="correction" id="">renormalise</span> so that it sums to unity.  This would give the<br>

probabilities for each candidate.<br>

                                                                                                                                                           
 <br>

What about something like the following for a non-random method.<br>

                                                                                                                                                           
 <br>

Prior to the election:<br>

                                                                                                                                                           
 <br>

Determine the probability of each candidate having a total vote<br>

within a% of the highest other candidate.<br>

                                                                                                                                                           
 <br>

If a is low, then this should be proportional to the probability<br>

of the candidate ending up in a tie.  It seems reasonable, but<br>

is that true ?  Anyway, call this probability <span class="correction" id="">Pc</span> for the <span class="correction" id="">cth</span><br>

candidate.<br>

                                                                                                                                                           
 <br>

Each voter votes a number of votes for each candidate and the<br>

candidate with the highest total wins.<br>

                                                                                                                                                           
 <br>

The benefit of giving a vote for a candidate is the probability<br>

of the candidate being in a tie times the utility of the candidate.<br>

                                                                                                                                                           
 <br>

Each vote cast by the voter should be weighted by:<br>

                                                                                                                                                           
 <br>

1/sum( <span class="correction" id="">Pc</span>*(votes cast for candidate c)^N )<br>

                                                                                                                                                           
 <br>

This means that votes for an candidate who is unlikely to win<br>

cost almost nothing (as <span class="correction" id="">Pc</span> is tiny), thus eliminating the spoiler effect.<br>

                                                                                                                                                           
 <br>

Using a different power than square/square root would mean that<br>

the probability of the highest utility candidate winning could<br>

be increased/decreased.  However, setting the power to high,<br>

would introduce <span class="correction" id="">alot</span> of noise to the system.<br>

                                                                                                                                                           
 <br>

The probabilities could be determined by the betting markets.  The<br>

problem would be a candidate with say, a 1 in 1000 chance of winning<br>

would give a noisy <span class="correction" id="">Pc</span>.  Making 'a' larger helps here but may reduce<br>

accuracy.<br>

<br>

It also means that voting for a sure winner is <span class="correction" id="">costless</span>.<br>

</div>

<div> </div>

<div style="clear: both;"><span class="correction" id="">Raphfrk</span><br>
--------------------<br>
Interesting site<br>
"what if anyone could modify the laws"<br>
<br>
www.<span class="correction" id="">wikocracy</span>.com</div>
<!-- end of AOLMsgPart_0_5b76a511-82f5-42af-bcd6-02ae6466279a -->


<div class="AOLPromoFooter">
<hr style="margin-top:10px;" />
<a href="http://pr.atwola.com/promoclk/100122638x1081283466x1074645346/aol?redir=http%3A%2F%2Fwww%2Eaim%2Ecom%2Ffun%2Fmail%2F" target="_blank"><b>Check Out the new free AIM(R) Mail</b></a> -- 2 GB of storage and industry-leading spam and email virus protection.<br />
</div>

</BODY></HTML>