depends on the game surely? All sounds very iffy to me this whole thing. Surely Dembo is still entitled to sue them for defamation/loss of earnings as the damage is already done now? If it was me I'd be going after them all guns blazing.

What is the model?  What are the relevant statistics?  What is the test by which these statistics are found significant or not significant?  I keep asking these questions, in vain.   

Further, would anyone be good enough to make public a set of data?

Also I have searched Prof. Regan's website, following the links supplied above, in vain for an answer to these questions.  Would someone care to point me to any material that explains all this?

@drogo: And how would any amount of such analysis as yours lead anyone to a conclusion?  In other words, so f--king what?

Quote from  http://www.cse.buffalo.edu/~regan/chess/fidelity/Varshavsky.html

Quote:


the bold highlight is my work

Markovich wrote on 09/20/10 at 19:22:05:


All the major website (ICC, playchess) will never disclose the methodology they used when banning suspected cheaters. What chess.com said is that, in their opinion, Yelena Dembo is a cheater. It remains a question who cares about this opinion. 

The games she played on that site can be reviewed and a reasonable player can judge by himself whether or not Dembo cheated. 

You seem to simply throw to the garbage the whole top-3 methodology. Well, I have bad news for you. Counting inaccuracies/blunders is an addition nobody mentioned. And, more importantly, the top 3 methodology was invented by this guy:
http://www.cse.buffalo.edu/~regan/
You are more than welcomed to ask him questions. But first, you should read (and understand) what he did. Here, have fun:
http://www.cse.buffalo.edu/~regan/chess/fidelity/

g-dog wrote on 09/20/10 at 21:53:40:


I am in such a position until you demonstrate some understanding of the statistical aspects of the problem and show how you modeled them.

g-dog wrote on 09/20/10 at 21:53:40:


Yes, and we would all just bow down before your alleged expertise if it were not for the necessity of bringing sufficient evidence before calling people cheaters.

Without any such evidence, it's rather difficult to tell the difference between people armed with sufficient data and good analyses, on the one hand, and miserable little piss-ants with chips on their shoulders, on the other.

g-dog wrote on 09/20/10 at 21:53:40:


What the exercise that I proposed would show would be the rate of exceedence ("cheating") resulting from random inputs.  It would show the chance that you had called Dembo a cheater by mistake.  Again, if you have no understanding of the statistical aspects of this problem, you have no business drawing conclusions, still less publishing them, on this subject.

Markovich wrote on 09/20/10 at 19:22:05:

Exactly. Labelling someone guilty without proof is defamation.

g-dog wrote on 09/20/10 at 21:38:42:

Sorry, but I will stop after this answer. I explained and gave a running example game.

I do run T3/T4 analysis reguarlarly but not always in my CC games on lss.chess-server.net and see what happens with my analysis without engine. Adding my statistical knowledge I know it's simply a sampling problem.

Check the game given in your databases: Vigus gives 17. - c5. I followed CA 11 and white made the first move new to me in with 21.Rad1. Combined with the demonstrated ignorance (no offense intended) of statistical inference this example shows clearly why it is easy to become a suspect without using an engine. Especially if you work on a repertoire.