Cheating in chess

[Jacob Ward]

I do not understand why there is such a resistance to organising OtB, to organising robust online events. Almost as if some players would rather see chess die and other players be deprived of the fun and experience, rather than see it played in a format they do not personally like.

I think what you are characterising as resistance is largely just people who don't find the prospect of playing online under those conditions appealing. Speaking for myself, I have no issue with other people playing under whatever conditions they fancy; I just personally don't see much attraction in playing in such regimented conditions. Then again I am in no danger of winning anything so I do not claim to speak for elite players trying to win significant tournaments, who may well accept or even want more stringent measures.

[NickFaulks]

There was nothing against her on audio or screen, some oddities on video, but she failed spectacularly the Regan test. Why? She beat 2 players 2100+ while having a rating of 1303 FIDE rapid.

This is the point we have reached?

[NickFaulks]

Then again I am in no danger of winning anything so I do not claim to speak for elite players trying to win significant tournaments, who may well accept or even want more stringent measures.

Grischuk has expressed himself with, as usual, impeccable logic.

He wishes to play, but will not do so while wearing something unpleasant on his face.
Other elite players may wish to play, but will perhaps do so only if he wears something unpleasant on his face.
They cannot play.

[Roger de Coverly]

This is the point we have reached?

There was an attempt a few years ago to have the Regan tests accepted in OTB chess in the absence of physical evidence. Particularly for juniors and particularly for intermittent forms of chess such as rapidplay, ratings can lack credibility. For someone who is a decent young player, beating a couple of players rated 2100 is well within the mathematics of plausibility.

[JustinHorton]

Interestingly, out of the 5 cases in the FFE tournament, the highest profile one (a winner) is still pending appeal. There was nothing against her on audio or screen, some oddities on video, but she failed spectacularly the Regan test. Why? She beat 2 players 2100+ while having a rating of 1303 FIDE rapid. Part of the appeal argues that her real level was more around 1800, and using this number in Regan's analysis reduces the anomaly level.
Worth noting that the appeal commission also thinks that Prof Regan's analysis on its own may not be sufficient proof, and that Master analysis of the game does not constitute additional analysis to Regan's software. Will be interesting to see what happens.

Well yes. I don't say so because I want to tear down Ken Regan, but I do want to find out what element of human judgement is going into using and interpreting his work.

[Matthew Turner]

This is the point we have reached?

There was an attempt a few years ago to have the Regan tests accepted in OTB chess in the absence of physical evidence. Particularly for juniors and particularly for intermittent forms of chess such as rapidplay, ratings can lack credibility. For someone who is a decent young player, beating a couple of players rated 2100 is well within the mathematics of plausibility.

The Regan tests have nothing to do with beating 2100 players, it is how well you match computer selections. Of course there has to be base case to test the performance against, so if you are assuming a player is 1300 and they are actually 1800 then this is going to be a very significant flaw. I would have thought in an individual case there was a way round this.
1. Take the player's last OTB games and perform a Regan test and measure their move matching performance.
2. Add on a sensible amount for improvement.
3. Perform the Regan test on the online games again using this new rating.

[Ian Thompson]

The Regan tests have nothing to do with beating 2100 players, it is how well you match computer selections. Of course there has to be base case to test the performance against, so if you are assuming a player is 1300 and they are actually 1800 then this is going to be a very significant flaw. I would have thought in an individual case there was a way round this.
1. Take the player's last OTB games and perform a Regan test and measure their move matching performance.
2. Add on a sensible amount for improvement.
3. Perform the Regan test on the online games again using this new rating.

I thought Regan said in one of his papers that he produced three sets of results using:

1. The player's current rating
2. The player's new rating, after the tournament
3. The player's TPR

It was then up to the tournament organiser to decide how to interpret these results.

That's not much different from what you are suggesting.

[Kevin Thurlow]

"1. Take the player's last OTB games and perform a Regan test and measure their move matching performance.
2. Add on a sensible amount for improvement.
3. Perform the Regan test on the online games again using this new rating."

Yes - I would have thought he would suggest that. Of course, it is always a problem providing advice, because all you can do is answer the question you are asked, based on the information given. You can ask for extra information, but might not get it.

[Matthew Turner]

The Regan tests have nothing to do with beating 2100 players, it is how well you match computer selections. Of course there has to be base case to test the performance against, so if you are assuming a player is 1300 and they are actually 1800 then this is going to be a very significant flaw. I would have thought in an individual case there was a way round this.
1. Take the player's last OTB games and perform a Regan test and measure their move matching performance.
2. Add on a sensible amount for improvement.
3. Perform the Regan test on the online games again using this new rating.

I thought Regan said in one of his papers that he produced three sets of results using:

1. The player's current rating
2. The player's new rating, after the tournament
3. The player's TPR

It was then up to the tournament organiser to decide how to interpret these results.

That's not much different from what you are suggesting.

Ian,
The Regan tests can produce quite a lot of pieces of information, but the two 'headline' figures are the Z score and IPR
The Z score measures how many standard deviations the performance is away from the expected performance
The IPR is a measure of what rating of player would be expected to have this performance (in terms of move matching).

So, in the French case we could compare the IPR with 1300 and then with 1800, but that isn't necessarily very scientific. If we have an accurate Z score then we can say something about probabilities, but that does require a reasonably accurate rating as an input.

[Roger de Coverly]

The Regan tests have nothing to do with beating 2100 players, it is how well you match computer selections.

The underlying problem being that the stronger the player, the more likely they are to find strong moves and the more likely it is these moves match those selected by an engine by coincidence.

Has this test ever been done? Take GM games from the pre-computer era. Run them through the Regan tests, but tell the test that the player ratings were 1500 or whatever. How many false positives would it find?

[JustinHorton]

I'm not at all sure that would be a sound procedure

[Matthew Turner]

The Regan tests have nothing to do with beating 2100 players, it is how well you match computer selections.

The underlying problem being that the stronger the player, the more likely they are to find strong moves and the more likely it is these moves match those selected by an engine by coincidence.

Has this test ever been done? Take GM games from the pre-computer era. Run them through the Regan tests, but tell the test that the player ratings were 1500 or whatever. How many false positives would it find?

Hundreds of thousands of players have been used to validate the Regan tests, that is how we know what to expect from a 1500 player. If a 1500 player is consistently performing at 2500 then either the initial rating is very very wrong or they are cheating. They might not be using a computer, they might be getting assistance in another way, but they are cheating.

[Roger de Coverly]

If a 1500 player is consistently performing at 2500 then either the initial rating is very very wrong or they are cheating.

How do you tell the difference, particularly for Juniors with out of date rapid ratings?

Actually why not forget about ratings determined by results at all? Produce ratings by analysing games if there's that much certainty about correlations between strength and matching to engine suggestions?

[MartinCarpenter]

If a 1500 player is consistently performing at 2500 then either the initial rating is very very wrong or they are cheating.

How do you tell the difference, particularly for Juniors with out of date rapid ratings?

Actually why not forget about ratings determined by results at all? Produce ratings by analysing games if there's that much certainty about correlations between strength and matching to engine suggestions?

You would need a sizeable sample of games that are 'known' to be clean to do that.

Which, for most people below a certain quite high standard, isn't something you'll have all that often. The rating isn't an ideal proxy for all those off database games but its definitely better than nothing.

[Matthew Turner]

If a 1500 player is consistently performing at 2500 then either the initial rating is very very wrong or they are cheating.

How do you tell the difference, particularly for Juniors with out of date rapid ratings?

You look at the evidence with an open mind don't you?

Some questions you might consider
1. Does the player have an alternative rating that might be more accurate
2. What does the player's rating progress look like
3. How many games is the rating based on