You are ignoring the author of this comment. Click to temporarily show the comment.
Yes. But there is less data on Seniors and Womens. I cover pre-2015 and post-2015.
July 22, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
And there I was thinking all these people were queuing to get into the Bridge events…

Someone will need to post a few photographs so that those not there can see what you mean.

The line must have been 200-300 people long. Combined they were wearing the same amount of clothing as one typical bridge player.

There is a secret way to get into the party from the Bridge event. I know this because I took my short cut and came back out into the main part of the Cosmo past the line and security checks. Security seemed very puzzled why I was headed the “wrong” way.
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Stefan: There are multiple algorithms, not just one. For example, looking for cheating on the opening lead is different than looking for cheating during the play of the hand.

I have tried to simplifying it by creating a single function that merges all the data. This makes it much easier for presentation purposes.

Assuming I have algorithms that generate a high value for cheating pairs, and a low value for honest pairs then I can state, for most algorithms,

1. If your value is high, this does not necessarily mean you are a cheating pair.

2. If you are a cheating pair, then your value is high.

When I look at multiple algorithms, then if you appear high in all/most of them, the conclusion is that you are a cheating pair.

Statistics in Bridge is complicated. You have to be able to look at the numbers behind the numbers.

I was showing some of the data to a World Class player a couple of days ago. I was showing how they compared to other world class players. Showing their weaknesses - relative to their peer group - in their game and their partners.

I was showing a table which ranked based on one of the cheating algorithms. They appeared high on the list. Not at the cheating level, just much higher than they might think. But, for this table, you need the data behind the data. There is a cross-check value I use. For this pair, they had played in some tournaments with weaker players. This will affect their “ratings”. This explained their ranking; so it was not a number associated with cheating. For the other pairs on the list, the cross-check value was very high, indicating that this result was through cheating.

The book has necessarily over-simplified presentation of the data.

If I am asked (by a Bridge organization, not a player), if someone is cheating, there are multiple factors to look at. Not just a single number. Of much more interest, and cannot be shown well in the book, is trend analysis. How/if players suddenly improved. I mention some of this in the book. I can detect when certain pairs started cheating, or more accurately, when their cheating methods became effective.
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Stefan: “”Forquet on lead against 2♠. What did he lead from ♠8 ♥K106532 ♦Q1092 ♣K2 and what do you think his partner had?"

Auction was 2 on his right by dealer. All pass. No-one vulnerable.

July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Tony: “Good to know my leading is bad enough to avert suspicion. I assume Robson is better than me, given the angle of your post Nic. I suspect someone who led trumps more often (not my style) would give less tricks away, but would it be effective in the long run. Anyway, well done on the research.”

As you could tell by the phrasing of the question…. but here's the answers from the data I have. It's Vugraph data because I need the opening lead:

1. When you play with Andrew, he finds a safe lead 83.3%, you are 81.2%. Both are above average for World Class players. Combined you are better than all the top ten pairs in the world except for pairs that have the same first initial in their last name.

2. When Andrew plays with Alexander, the roles are reversed. Alexander is 81.3%, Andrew is 77.8%. Alexander is above average for WC players. In this partnership Andrew is below average. I can give him some lessons if he wants.

Perhaps a little unfair to mention A/A, but they are presumably a reference pair you know. The data says Andrew is a phenomenal player (page 151) - this is from data that is nothing to do with cheating.

But…. a “safe” lead is not the best way to measure the skill level of a Bridge player in team games. At least in my opinion. There are better ways. I'll let you figure out what they might be; some of them are in the book.

You asked about trump leads. Page 132 in the book. Data from top 100 pairs. Sorry, I didn't list all 100 pairs.

Your perception is, “I suspect someone who led trumps more often (not my style) would give less tricks away, but would it be effective in the long run”

The computer says…

You/Andrew rank #17 out of the top 100 pairs for leading trumps. You are ranked 99/200 players for leading trumps - 10% of the time you will lead a trump. You may not think it is your style, but you are just above the median for top players leading trumps. Andrew (14.5%) ranks #15 when playing with you.
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.

1) how was the analytics/code validated?

Tough question. How can you validate it? Only way is to see if it detects the known cheaters. See Chapter 1 of the book. There is a snippet here: http://www.detectingcheatinginbridge.com/selections.html but it does not have all the chapter.

I show what happens when one of the formulas is applied to the data and it lists seven pairs from the top 66 pairs with the most data. Six of the pairs are known cheaters. From the book, “The statistical chance of a random formula placing these six pairs in the top seven out of a selection of 66 is 1 in 13 million.” The other pair is discussed. Their data is shown in the book. Draw your own conclusions. They defend better than B/Z over a large number of boards.

For those with software background: for some of the early work where I needed to double check the results, I would write two complete different routines. Or do one in a programming language and also process the results in Excel to make sure the same results were being generated. This helps catch bugs in the data analytics code. I still does this at times, particularly if new or important code is added.

2) does the analysis have the power to separate signal from noise with acceptable confidence? Even when data can be highly sparse due to sporadic behavior and avoiding glaringly obvious abuse?

Yes. I think it does. BUT, you must have a large sample. There is a table of contents in the snippets URL above. I cover this, and other errors in the underlying data, in Chapter 12.

Everything works on the concept of the Law of Large Numbers (LLN). Chapter 7.

3) how does the analysis deal with correlation and cause? Issues like correlation to a third variable, reverse correlation, and other biases must be clearly avoided.

Beyond the level of readership of the book!

If you are in Las Vegas, I'll explain personally; but too long for a BW post.

If you cheat, you are better on defense than your peers. This can be measured. That is the short answer.

4) how does the analysis avoid confirmation bias?

I use all data, not selected data. This is the most common problem with presentations on cheating, just using selected hands. I don't try and explain away false/positives.

To give an example: There are chapters on Opening Leads - 29, 30, 31

I pose the question: how would you detect cheating on the opening lead?

5) how does the analysis identify superior skill, analysis and methods at the table. That is, how is genuine skill differentiated from illicit behavior? Can the same be said for blunders by opponents?

Take the best non-cheating pair in the world, your opinion not mine. Look at the data behind them. Compare to cheating pairs.

I show that the top pairs, with large amounts of data, clump together on a scatter plot. You can only be so good. You will make mistakes. I know the mistakes frequency of top pairs. I know that top pairs will make a bad lead 19% of the time. I can show you data on pairs that make bad leads 15% of the time. Do they cheat or are they better Bridge players? Answer is in the book.

And…. I know your data. I can show where you are relative to the top players/pairs.

The question deserves a little more respect as it is possible to distinguish between your mistakes and your opponents blunders. I show how this is done. This is one of the more sophisticated checks in the book. See ACDF3 on page 51.

“how is genuine skill differentiated from illicit behavior”. Page 146 describes a test on data from before 2015 and after 2015. Before 2015, there are ten pairs (includes five known cheating pairs) from the top 150 pairs with the most data that had a “skill rating” above a certain level. Post 2015, for pairs with the same amount of data, there are none.

It is comparative type tests like this that show cheating.

I make some statements about the known cheating pairs about areas of the game they are cheating in, above and beyond what is known. Someone will eventually find the players codes from the videos to prove these statements and validate the work.
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
The booksellers in Las Vegas have it or http://www.detectingcheatinginbridge.com (US shipping only)
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Stefan. True story. The original working title was “Detecting Cheating in Bridge Using Statistics and other Bridge Statistics”.

But then I made up a statistic that whenever you mention the word “Statistics” in a book title, sales drop by 90%. Mentioning it twice drops it even further.

In the end market research said people won't buy/read a book with the word “Statistics” in it and the reviewers recommended a name change. “Statistics” reminds everyone too much of school and work was the comment.

Also the original book was really really wide to fit the title in so we had to change it.
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@iRichard: “Nicolas, don't you have data for Forquet-Garozzo?”
Yes. Forgot to include them in the BW post.

I don't have much - 190 boards. Not quite enough to be statistically significant for some of the algorithms, but enough for a couple of them. They rank #516 on the amount of data I have on pairs.

Here's a Forquet opening lead question from the book (p 185). I won't give the answer as don't want to give out spoilers. Forquet on lead against 2. What did he lead from 8 K106532 Q1092 K2 and what do you think his partner had?

Chapter 46 goes over the data from the Bermuda Bowls 1955-1991. Forquet is mentioned with three different partners: Guglielmo Siniscalco/Pietro Forquet, Benito Bianchi/Pietro Forquet, Benito Garozzo/Pietro Forquet. There are ten pairs from that era with enough data to be analyzed.
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Barry: Thanks for the input, I particularly appreciate the price!.

The presentation of Figure 15 on the web site is missing annotation of the axes in the figure, but is described in the surrounding text. My son helped me put the web site together so it is a grab from the book, paste to PNG, web resize, stuck in HTML. The axes are more fully described in the book. The explanation value for the X axis takes a chapter and half! The concept that it is possible to calculate this value is at the heart of the book. Too much for the web site.

As I am sure you know, formatting/presentation is hard. Most books are 6“ x 9”. There are some tables that I wanted to put it that could not be made thinner without removing necessary content, so I ended up with a 7x10 book. Reducing font size, impossible to understand table headers were all tried, but the easiest solution was to move to a 7x10 book. Given a 7" width book, and with padding, margins etc. there is still little room. The first format for the reviewers was the typical letter size Word document. For pictures/figures there is a problem of putting too much information in the diagram. In nearly all cases the axes description is in the body, not the figure. Part of that is where the figure came from, as I use different tools to generate the data. Labeling the Y axis is particularly hard within the space constraints.

There are always tradeoffs when putting a book together.
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
I use different data sets. I use either data from Vugraph, or from the EBL/WBF/ACBL web site, or The Vugraph Project. I am adding data from another NBO.

The software is both “sophisticated”, but also “simple”. Same with the algorithms.

There is not one magic formula to detect cheating. I have different approaches to detect cheating on bidding, leads, defense.

To keep things simple, I do have a single formula that condenses everything to one value. It is a “mistakes function”. A typical player will make similar numbers of mistakes on defense and declarer play. A cheating pair makes fewer mistakes on defense. This is the MF value.

I can calculate at the tournament level. For a given tournament, similar players, I expect this value to be an approximate constant given a large number of boards.

When I remove the data from cheating pairs, I am removing just their data. The formula is designed to detect cheating. The value is expect to go up. It does in all cases with large data sets. Put simply, cheating pairs make fewer mistakes on defense and this can be detected.

I realize that this concept is new. Very few people understand it at the moment.

If you accept that such a formula exists and the “quality” of a bridge tournament can be measured, why does the data show such a big change from pre-2015 to post-2015?

I show that the formula works by applying the data to 11,000,000 ACBL records. As you get better. You make fewer mistakes. I can calculate the value for players above a certain masterpoint level.

The data shows there are at least three different eras. 1955-1983. Up to 2015. Post 2015.

The results are consistent for major tournaments within those eras.

What else changed in Bridge?
July 21, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Avon: I don't have much data on Eugenio Chiaradia and Massimo D'Alelio. They rank #548 based on the amount of data I have on each pair. As such, their results may be subject to more statistical variation than others.

I was to put this pair in the top 200 pairs, then Chiaradia would rank #315. D'Alelio in this partnership was awful. He would rank #399. This is based on “safe” opening lead. Based on this statistic, you would presume that they are bad players and do not cheat. But “safe” leads are not a good indicator of cheating; there are better ones (one more plug for the book - details are in it).

Part of detecting cheating is remaining ahead of the cheating pairs. As such, there is quite a bit of information I did not put in the book. I have a formula that calculates “dishonesty”. It is not in the book. It shows me the difference between their declarer ability and their defensive ability. I can then compare with other top pairs. A high dishonesty rating does not mean you are a cheating pair. All cheating pairs have a high dishonesty rating.

There is not much data on Chiaradia/D'Alelio. But I can generate their “dishonesty” rating and form an opinion based on this and other data on the likelihood they were cheating.
July 20, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Tony: “It is dangerous to get too excited about opening leads, as a passive leader will ”find“ partner more often than an aggressive one. With or without help.”

Absolutely (dangerous). I cover the different styles of opening leads in the book. I was chatting to Bob Hamman yesterday. He has two different styles with this three main partners - Zia, Paul Soloway, Bobby Wolff. He is able to mimic his partner's style.

So…. I have data on you/Andrew Robson (I have data on everyone!). Which of you is the “better” leader. Let me define “better” as the person who makes more “safe” leads (ones which do not give up a trick).

You rate #170 on the list of the top 400 in terms of safe leads. Do you think Andrew rates higher or lower when playing with you? Actually this is a bit of a spoiler because this result is listed in the book (page 113).

Which of you (Tony/Andrew) is the more “aggressive” leader? One who hits partner's honors more often?

I also have data when Andrew plays with Alexander Allfrey. Alexander ranks #162. Do you think Andrew rates higher or lower when playing with Alexander? (Result not in the book, but will give it).

Part of the book is handling perception/reality.
July 20, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Avon: “Nicolas: We were certainly looking at different things.

I searched for deals where Blue Team players made a lead that:
- almost no strong player would choose
- suited partner very well indeed”

Absolutely. But you are now getting into subjective opinion on what strong players would do. Some people make quirky leads.

The software can look at ALL leads and see how successful they are.

This is different from finding opening leads that are “suspect”.

I cover this in the book.

A good thought exercise (before you read the book please!) is to construct an algorithm that you think will detect cheating on the opening lead. See what you end up with. Then compare to what is in the book.
July 20, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Richard, “Not that much difference between Avarelli and Belladonna when playing with each other. The difference between Belladonna partnering Avarelli and partnering Garozzo is eye-catching.

How was Garozzo partnering (a) Belladonna, (b) someone else?”

In order for statistical analysis to be valid, you need a large data set. I only have enough data on Belladonna/Avarelli and Belladonna/Garozzo so can't comment on Garozzo with anyone else.

The difference is not that eye-catching. Better partnerships understand each other's overcall style etc.
July 20, 2019
Nicolas Hammond edited this comment July 20, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.

Obviously, I deliberately chose a famous player/pair close to them in the rankings.

Detecting cheating on opening leads is quite complicated.

One possible approach is to check for how successful the lead, which is the data I presented above for some pairs. Is that the best approach for cheating? Details in the book. These details are very useful for improving your opening leads.

Meckstroth ranks #222 out of the 400 pairs when playing with Eric.

Zia ranks #18 when playing with Meckstroth.
July 20, 2019
Nicolas Hammond edited this comment July 20, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
@Avon: I took the data from the top 200 pairs from the top tournaments. This includes Belladonna-Avarelli.

I sorted the 400 players based on the percentage of safe leads, i.e. leads that did not give up a trick.

Avarelli ranked #268, two spots above Zia, playing with Michael Rosenberg.

Belladonna, playing with Garozzo,, ranked #33, two spots below Bob Hamman, playing with Bart Bramley.

Belladonna, playing with Avarelli,, ranked #335, seven spots ahead of Jeff Meckstroth, playing with Zia.

Avarelli was not that good at opening leads…. using this statistic.
July 20, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
I met him as well. He has a PhD from MIT. He's still working on new mathematics.
July 19, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
See http://www.detectingcheatinginbridge.com for details about the book, include from snippets from some chapters and some of the figures in the book.

“Next: If I understand things correctly, it should now be possible to inspect the hands that WILL be used for individual tournaments, identify boards that the expert system will use, and then generate a hypothesis in advance regarding pairs that are unusually likely to make the right decision.” Not quite. What has happened is that the software identified a pair. Video of this suspicious pair was done, the software analyzed the data, and was able to predict boards where something “unusual” happened before the video was watched. Watching video proved that indeed something “unusual” happened on those boards.

In order for this system to work it relies on a large data set. Given a large data set, if a pair is defending better than say, Balicki/Zmudzinski who were transmitting information about suits, then this raises flags.

There are multiple algorithms detecting cheating in different parts of the game. If they are released, then cheating pairs will modify how they play in certain parts of the game making it more difficult to detect cheating. Comparing results from top players quickly flags the outliers. Once the outliers are outside normal statistical deviations, they become more suspect.

However…. I hope the long term effect will not be in detecting cheating but how using statistics can improve your own game.
July 19, 2019
You are ignoring the author of this comment. Click to temporarily show the comment.
Microsoft Ready and Inspire 2019 convention (40K attendance) finishes today. Presume it might free up a certain person to come and play.
July 18, 2019
.

Bottom Home Top