If you’re a fan of statistical analysis of sports, or have even a passing interest, I would highly recommend checking out Tom Tango’s blog. He’s mostly a baseball guy, but also comments on other sports, and does so with a statistically-minded viewpoint. More than anything, I like his ability to simplify concepts, and explain the theory behind why statistics are effective at making an argument (or ineffective). With Tango, you don’t get lengthy lectures about statistical methods with just a paragraph at the end explaining what it all means. I’ve found that when he has something to say, it’s usually with a very clear, simple approach that is easy to understand and apply.
With that said, this post in particular about the voting for the NL Cy Young Award caught my interest. This was because of his anecdote in the middle of the post:
This reminds me when I did that football pool when I was in college, where you rank winners from 14 to 1, and you get points based on where you rank them. While my friends sweated every week to figure out the order, spending up to an hour, I simply went with Vegas and called it a day. I went into the last week in 2nd place out of 22.
Tango explains that he, as an individual, had no knowledge to add above and beyond that of the collective wisdom, which goes into determining the point spreads and betting lines for a particular game.
It reminded me of my work with SILVA, and trying to predict who would win particular fights. Those who have been reading my blog know that every time I write a post summarizing SILVA’s predictions for a particular UFC event, I include a feature called “the big underdog pick of the week.” Essentially, I mention which predictions SILVA makes that are also betting underdogs. If you’ve been paying close attention, you probably realize that SILVA’s track record making such picks is extremely poor.
At UFC 139, for example, SILVA made two underdog picks: Rafael dos Anjos over Gleison Tibau and Brian Bowles over Urijah Faber. The dos Anjos pick is one I can live with – he was a narrow underdog and SILVA only favored him by a very slight margin. The Bowles pick, on the other hand, really ended up being a disaster, as Urijah Faber was not only a strong favorite, he ended up beating Bowles in very decisive fashion. When SILVA picked Bowles to win by a 31-point gap, I went into the data to try to prove that SILVA was making a bad pick. As it turns out, Bowles’s recent record was just plain stronger, and his and Faber’s data on Fight Metric was very comparable. I couldn’t prove SILVA wrong, so I went with Bowles. As it turns out, Faber was the winner and it wasn’t close. It didn’t matter what data I used… I wasn’t going to have a better idea than the “collective wisdom” about who the winner would be.
Overall, SILVA 1.1 is 1-4 when it predicts an underdog to win. The one triumph was actually Junior dos Santos over Cain Velasquez, but SILVA missed on Rob Broughton at UFC 138, Pablo Garza at UFC on Fox, and Brian Bowles and Rafael dos Anjos at UFC 139. It’s an extremely small sample of fights, so there’s not enough here to make any conclusions, but the question to be asked is: is it even worth looking at SILVA to predict who will win when there are betting lines available to look at?
I think SILVA still has value because of the ability to look at a list of fighters in a specific weight class. Sure, we may know that Chris Weidman is a 4-1 favorite against Tom Lawlor, but what would the betting lines be if Weidman was to fight somebody like Chael Sonnen? SILVA may not be as accurate as the betting lines at making predictions, but it still provides a decent estimate (most of the time), and provides for the ability to look at the hierarchy of where Weidman rates compared to his peers (he rated 15th at middleweight before beating Lawlor; I have yet to update the SILVA scores following UFC 139).
As for making predictions, I can’t recommend SILVA over simply looking at the betting lines for a fight. SILVA’s track record just is not strong enough to do so.