[EM] Round robin tournament statistics

[Kristofer Munsterhjelm]

fsimmons at pcc.edu wrote:
> ----- Original Message -----
> From: Kristofer Munsterhjelm

>> I also imagine it would be useful in places where it's hard to 
>> strategize or the context means there won't be any strategy. Such 
>> examples might be computers in a redundant system voting about an 
>> observation under uncertainty (the "strategy" will be a random 
>> distortion) or actual round robin tournaments (where engineering a 
>> Condorcet cycle based on just one's own matchups would be quite
>> hard).
> 
> 
> Or how about in the context of ranking websites for search engine hits?
> 

There are strategists in the realm of website ranking: they're called 
SEO companies. If I remember correctly, Google uses some form of local 
Kemenization (bubble-sorting, basically) to harden its eigenvector 
method against gaming.

If we assume Google's ballot model, where every site votes for every 
site it links to, then we have, in effect, a bunch of really truncated 
two-level ballots. Strategists can't bury pages (other than by not 
linking to them), but they can strategically choose which pages to link 
to in the first place. This has an advantage in the eigenvector system 
because linking to a page makes it more authoritative - its voting 
weight is related to the number of pages (weighted) that link to it.

The strategy is there, but it's different than in other voting systems. 
Some trust networks (but not Google's, AFAIK), have manipulation 
resistance proofs, e.g. http://www.advogato.org/trust-metric.html .