26 comments

[ 3.5 ms ] story [ 69.4 ms ] thread
Great work! As someone who never does data exploratino or use python, I appreciated how annotated your iPython notebook was. It was fun to read through. (link https://github.com/BuzzFeedNews/2016-01-tennis-betting-analy... )
Very clever. I just hope Buzzfeed don't chase you down and force you to take the list down!
Something feels a bit ethically questionable about publishing the names like this.
Which is why buzzfeed didn't do it. I'd rather have had this come out because buzzfeed turned the data over to ITF & ITF took action against the players.
It also seems really stupid. I would not be surprised if he got threatened with lawsuits from the named players.
On what grounds? They're not making any claims, they're just noting the fact that interesting and unlikely things happened to these guys' odds (public data) during matches.

They've gone through a reasonable degree of trouble to make sure their numbers are completely right too.

The original authors published accurate data. The medium author accused named individuals of cheating.
They're not, read again, in fact they start with a disclaimer explicitly stating they are not commenting on the allegations, only de-anonymising the data.
I'm not necessarily saying it's illegal, or unethical. I just think it would be a bad idea. It looks like they have done decent work and have not accused players of cheating. But the issue is making international news, so I'm definitely assuming some named players are talking with lawyers about how to deal with this. And I personally would not want to deal with those lawyers, even if I had done nothing illegal or wrong.
I find this to be a useful example of the dangers of so-called anonymized data (namely, the danger that it isn't).

I'd be curious to know how much fuzzing of the percentages published by Buzzfeed would've been sufficient to prevent this sort of obvious matching.

It looks like someone is going to deal with a lawsuit. Good work, but if I was one of those players, I'd be very mad.
Not very likely in American courts given the public figure libel standard.[0]

Essentially, in order to accuse a public figure (e.g. politicians), you have to prove that the defendant acted with actual malice.

Actual malice in this case is defined as either actual knowledge that the statements were false, or reckless disregard for the truth.

A reckless disregard for the truth is more than mere negligence, the person making the statement must have actual and demonstrable doubt as to the truth or falsity of the statements made. [1]

It is very difficult to meet that standard.

[0] https://en.wikipedia.org/wiki/Public_figure

[1] New York Times Co. v. Sullivan (1964)

I have no problem believing that there is match-fixing in tennis, especially among the non-elite ranks. However, it is still an extraordinary claim, and as such, requires extraordinary evidence. They have examined a dataset that includes match losses by 1,509 unique players. Among others, they have accused a fellow whose match results, by their own estimation, are 4.7% likely to have occurred by chance. If you're going to be involved in throwing around accusations of this magnitude, I'd like a little more rigor in the statistics. I don't know how they sleep at night.
Yep. "Assuming the initial odds are right, these guys may have cheated". That's one hell of an assumption. The odds moving is one of the strongest indicators of the initial odds not being right. Market efficiency etc.
Is the market truly efficient? Value betting is all about finding when it's not.
Including the 4.7% in the list really stands out. Given that the number of professional tennis players in the world exceeds 100/4.7 that sure looks like it should have been rejected due to the green jelly bean fallacy. [1]

[1] https://xkcd.com/882/

Yup, looking at these stats puts a different complexion on the attitude of the tennis authorities (which in the case of the Davydenko-Arguello match screamed "cover up")

If someone announces to you that they've done some statistical analysis of 1500 players and they were shocked to find that a few of them of them fail a suspicion test at a 95% confidence threshold, you're going to question their grasp of statistics at least as much as the players' integrity.

And that's before you start considering that the "suspicious movement" flag they've chosen is a starting point for investigation by bookmakers rather than a sure indication of something being probably wrong, and their methodology for correcting outlying mispriced odds which differ from other bookmakers is laughably inadequate. Have a bookmaker whose opening price is 9.99 points longer than the average for the other bookies whose price correction moves it to 0.01 shorter than the other bookies? That's "suspicious", apparently.

And that's just looking at it from a pure statistics perspective and not assessing context which might exonerate them.

One of the listed names is a Grand Slam winner who [coincidentally] played his last single game this week to a rapturous crowd, and earned >$20M in prize money over the course of the career. It's possible he augmented that by throwing the odd match for 100k, but far more plausible his odds moving all over the place were the result of his form and injuries being all over the place in the latter part of his career.

(For the more forensic approach into Lleyton Hewitt's "suspicious" odds movements, someone wrote this: http://www.sportdw.com/2016/01/tennis-fixing-buzzfeed-hewitt... )

I think you are significantly overstating the extraordinariness of these claims. People are entitled to their own priors, I suppose, but if someone asked "What is the probability that there is at least one player that has taken money to lose a match in Tennis" if you answered something less than 1% you are probably being far too trusting. It doesn't take that much evidence to turn 1% probability to 80% probability. I also don't understand why you're focusing on the 4.7% player. You have to set the cutoff somewhere, and it's not like they specifically named Michael Russell. The accusations in this case do not have much magnitude. People are accused of cheating all the time, and the evidence in this case is as strong as in other cases.

Edit: On further thought, you are probably saying that the people who published this Medium post are being irresponsible publishing the names. That's a little different than I was thinking, and more reasonable I think. That said, I don't think Michael Russell is in much risk here professionally, unless this statistical analysis is much more iron-clad than I expect it to be.

(comment deleted)
What is the true motivation of doing this? (publishing the names, that is)
Tennis is a puzzling sport - you can be up 5:0 and yet lose most of the following games as well as the match itself. Anyone that plays tennis knows about this "roller-coaster" (especially visible in WTA). "Unusual" is quite usual in tennis, honestly. You should take it into an account when you do stats. Obviously, people try to exploit this fact to make believable pre-determined matches, but it's really difficult to tell who is doing it, and who is simply more "unstable".
Seems choking is also now cheating.

Tennis is a confidence game, miss a big few shots and you stop going for them. It's all downhill from there.

We need to see the phone records.