<div> Jobst Heitzig wrote:<br>

> Now, I have the impression that a slight modification of the tax
formula may reduce this incentive considerably. Consider this tax:<br>

><br>

>   sum { R(W,k) - R(W(i),k)<br>

>   + sum { ( R(W(i),k) - R(W(i,j),k) ) / 2<br>

>   : j different from i and k }<br>

>   : k different from i }<br>

><br>

> where W(i,j) is the winner after removal of both i and j. If I'm right, this formula makes it ineffective to misrepresent ratings for both individual voters and pairs of voters.<br>

><br>

> Please check this!<br>

                                                                                                                                                            
<br>

This is a 2-d sum, so people could pay way more than they thought.<br>

                                                                                                                                                            
<br>

The outer loop just calculates the Clarke tax<font face="Arial, Helvetica, sans-serif"> and the inner loop is<br>

an adder</font>.<br>

                                                                                                                                                            
<br>

Anyway, in simple terms, the system is (I think)<br>

                                                                                                                                                            
<br>

- count up the totals excluding your ballot<br>

- If the result is different from the actual result<br>

  pay to change it to the actual winner<br>

  (i.e. the amount the actual winner lost by)<br>

                                                                                                                                                            
<br>

Your system adds<br>

- If the result is <font face="Arial, Helvetica, sans-serif">further</font> changed by removal of another ballot<br>

 in addition to yours, pay half the cost to switch it to<br>

 the winner that would result with just the removal of your ballot<br>

                                                                                                                                                            
<br>

The 2nd cost gets added for every other potential pair that your ballot<br>

can match with.  In a large election the tax could be massive.<br>

                                                                                                                                                            
<br>

For example, if lots of people cast<br>

                                                                                                                                                            
<br>

A) 10<br>

B) 0<br>

                                                                                                                                                            
<br>

and A defeated B by 15,<br>

                                                                                                                                                            
<br>

then the payment would be huge.<br>

                                                                                                                                                            
<br>

The removal of any one voter's ballot would have no effect, A would still<br>

win by 5.  However, the removal of any 2 would cause B to win.<br>

                                                                                                                                                            
<br>

If there were 1000 voters of that type, then each of them would have to pay<br>

999 times 2.5 in the tax (A loses by 5 999 times and they pay half each<br>

time).<br>

<font face="Arial, Helvetica, sans-serif"><br>

My preference for a solution here is to limit the size that people may bid.<br>

For example, the highest 10% of ratings are assumed to be equal to<br>

the lowest of them.<br>

<br>

I.e. where (Rmax-Rmin) on a ballot is greater than the 90th percentile,<br>

rescale the ballot by <br>

(90th percentile)/(Rmax-Rmin)<br>

</font></div>

<div> <font face="Arial, Helvetica, sans-serif"><br>

This means that bidding a really large number will just be ignored and your<br>

ballot rescaled.<br>

<br>

This does break the optimal utility effect though.<br>

<br>

</font></div>

<div id="sig205" style="clear: both;"><font>Raphfrk<font face="Arial, Helvetica, sans-serif"><br>

</font></font></div>

<div> <!-- end of AOLMsgPart_0_423adf18-83bc-4f7a-ac72-fabc8a334b70 -->

</div>
<div id='u8CACCF0E05FDE33-116C-F50' class='aol_ad_footer'><FONT style="color: black; font: normal 10pt ARIAL, SAN-SERIF;"><HR style="MARGIN-TOP: 10px">It's time to go back to school! Get the latest trends and gadgets that make the grade on <A title="http://shopping.aol.com/back-to-school?ncid=aolins00050000000007" href="http://shopping.aol.com/back-to-school?ncid=aolins00050000000007" target="_blank">AOL Shopping</A>.</FONT> </div>