NBA playoff predictions 12/06/17

Match-Up Naive Bayes Neural Network KenPom Outcome
Cleveland Cavaliers – Golden State Warriors Golden State Warriors (0.943) Golden State Warriors (0.881) Golden State Warriors (0.5776) Golden State
Accuracy 1/1 1/1 1/1
Postseason accuracy 50/79 50/79 50/79

NBA playoff predictions 07/06/17

Match-Up Naive Bayes Neural Network KenPom Outcome
Cleveland Cavaliers – Golden State Warriors Golden State Warriors (0.903) Cleveland Cavaliers (0.553) Golden State Warriors (0.5171) Golden State
Accuracy 1/1 0/1 1/1
Postseason accuracy 49/77 48/77 49/77

Edit: a lot of the earlier rounds were not very exciting and now the finals risk becoming boring as well. Theoretically, there could be 8*7 + 4*7 + 2*7 + 7 = 105 matches in a post-season. I still wonder San Antonio with Leonard could have changed things.

Professional sports analytics

Ars Technica ran a very interesting article about how Benfica uses sports analytics to buy players, develop them, and sell them at a profit:

Let me give you a few examples. Benfica signed 17-year-old Jan Oblak in 2010 for €1.7 million; in 2014, as he blossomed into one of the best goalies in the world, Atlético Madrid picked him up for a cool €16 million. In 2007 David Luiz joined Benfica for €1.5 million; just four years later, Luiz was traded to Chelsea for €25 million and player Nemanja Matic. Then, three years after that, Matic returned to Chelsea for another €25 million. All told, S.L. Benfica raised more than £270 million (€320m) from player transfers over the last six years.

Those guys are doing things professionally, and with significant stakes, that academic sports analytics researchers do as well (and often have done as proof-of-context before). The article lays out what kind of different data sources the club uses, how to integrate and process them but when I read something like:

Data science and machine learning is a new discipline, and it’s not easy to find people who know how to use these tools, and how to integrate them with our day-to-day work.

I can’t help but chuckle. Precisely the “disciplines” of machine learning and data science (pdf) are very obviously not new. They do have filtered into the wider conscience only recently, however, in the same way that for some, “machine learning” is equivalent to “deep learning” because this is the most-talked-about ML technique in use today.

When the gatekeepers are biased, science has a problem

There’s been a paper that claims that lesbian desire exists in the human population because heterosexual men appreciate it. Or, in the words of the paper’s abstract:

Accordingly, this paper proposed a theoretical framework where, during the period of human evolution, same-sex attractions in women were under positive selection. The source of positive selection has been male preferences for opposite-sex sex partners who experienced same-sex attractions. This theoretical framework was used to generate four predictions that were tested in two online studies which employed a total of 1509 heterosexual participants.

So they asked self-identified heterosexual men and women! And found:

It was found that heterosexual women did not desire partners who experienced same-sex attractions, but a considerable proportion of heterosexual men desired partners who experienced same-sex attractions. In addition, it was found that men were more sexually excited than women by the same-sex infidelity of their partners, and they desired more than women, their opposite-sex partners to have sex with same-sex individuals.

There’s a bunch of things to be said about their hypothesis, setup, and results but a very important one is mentioned in the article linked above:

Diana Fleischman, a psychologist at the University of Portsmouth said, the paper is showing the influence of porn on men’s preferences as supposed to actual science.

“The paper totally ignores a lot of other possible hypotheses and makes claims that are really not supported by the evidence they provide.

Two women having sex with one man is such a common theme in pornography that I think it is very difficult to parse out that particular variable.

There’s a big cultural influence of porn because men are more likely to form associations through classical conditioning and stimulation and sexual arousal.”

What bothers me even more than that someone did this study in the first place is that it got accepted! Personality and Individual Differences is an Elsevier journal (which might not mean much but at least indicate that this is not a “pay-to-be-published”-mill) with an impact factor of around two (which also might not mean much apart from the fact that it indicates that this thing has been around for a while). And there was no reviewer that brought up this point about the conditioning by porn?!