[Foundation-l] Raw data of 2009 Board election ballots

Gregory Maxwell gmaxwell at gmail.com
Wed Aug 26 15:09:50 UTC 2009


On Wed, Aug 26, 2009 at 10:12 AM, Tim Starling<tstarling at wikimedia.org> wrote:
> Let me say for the record that I'm not at all happy with this data
> being released, since it allows vote-buying. Even if the numbers given

Although I was trying to avoid advertising it in public this was
something I'm aware of and had pointed out to the election committee,
but something I don't consider to be a risk we can meaningfully
address by not releasing ballots.

Quoting myself from a private email:

  I think the bigger risk is vote watermarking leading to vote buying:
  I.e. You could register with my site and tell me you want to vote for
  "M,ABFO,CDEGHIJKLN"  I then tell you I'll give you $10 if someone
  votes for "G,M,ABFO,CJ,LN,DEGHIK".   I make sure not to give out the
  same modified ballot twice, and I pay people if the ballots end up in
  the report.
  To fight against this I recommended that the WMF delay ballot
  disclosures for a few months and announce that they'd be doing so.
  People will be less inclined to wait for their $10. ;) I don't think
  stronger protection is justified because people could just load some
  toolbar that votes for them like subvertandprofit uses.

http://subvertandprofit.com/content/prices is a good cluestick for
people who think you can solve quality challenges with voting. :)

So, basically, my position is that the risk of buying due to vote
marking isn't much greater than the risk of buying based on the puppet
voter intentionally using a buyer controlled web-browser to vote...
and that we can equalize the risk by simply delaying the ballot
release a little bit, but not so much as to degrade the value of the
ballots as evidence that the election was conducted fairly.

> by voters are reduced to the smallest values which still give the same
> rankings, with 18 candidates there are 18 factorial possible
> orderings. That number is sufficiently higher than the number of
> voters that a party wishing to buy votes can specify a voter-specific
[snip]

Nitpicking, but the number of possible unique ballots is much greater
than the factorial because of equality, and equality must be preserved
in order produce the election calculations. The formula mostly easily
represented is a messy multipart recursive formula, which I'll spare
you (in part because I don't know that I have all the boundary
conditions right).  It's less than X!*2^(X-1).




More information about the wikimedia-l mailing list