Pages

Thursday, January 31, 2013

Curse You, Nate Silver

OK, not really. I am a big fan of the Silver one, and his application of quantfied reasoning to realms rife with superstition.

But his Super Bowl Analysis got me wondering about the data. He uses a stat called SRS, Simple Ranking System, from pro-football-reference.com. They have a nice linear algebra explanation of it on their blog. Mr. Silver used this stat to consider whether good defense or good offense was more conducive to winning. By looking at the top 20 in each category, he concludes defense is the key. Or more of a key than offense.

Always fabulous and frequently mathy Foxtrot.
I got to wondering about the differences. Maybe what mattered was your offense to opponents defense, or vice versa. I started putting his data into a spreadsheet, but because teams are rarely exceptionally good at both (exceptions 68 Colts, 96 Packers), there wasn't enough data. I was surprised to find that Pro Football Reference is free and an awesome resource for data. So I started filling in a spreadsheet, figured I'd tumblr it and people could fill in the rest if they were interested.

I got hooked, which is why I say, "Curse you, Nate Silver."

Turns out my idea about the differences was only mildy predictive. I took NFC Offense - AFC Defense and compared to AFC Offense - NFC Defense. ("How is that different from NFC Total SRS  - AFC Total SRS?" I would ask students if I had them this semester.) Here's the Google spreadsheet.


There's definitely some things to notice there. My next thoughts were to compare defenses, offenses, and finally to look where those two agree.


Bad news for Baltimore. When a team is favored in defense and offense, they've won about 3/4 of the time. So Raven fans you should be getting 3:1 odds when most people are feeling it's pretty close. Looking at these numbers, you could consider it the 4th or 5th biggest upset should Baltimore win. No 68 Jets or 07 Giants, but significant.

What do you notice about the data?

No comments:

Post a Comment