Thursday, June 15, 2017

Sports Modeling: Long Answer to a Short Question

The sacred shield of the NFL.
My buddy Joe and I have been planning a shared project for later this summer, a combined preview of the coming NFL season.  Regular readers and/or USMA classmates will hopefully remember Joe's blog, A Hoosier on the Potomac.  As of this writing, the plan is to alternate previewing divisions over the course of four weeks, doing two divisions per week.  As a Giants' fan, I've drawn the NFC.  Joe, a Colts fan, is doing the AFC.

We spent most of our recent exchanges talking about formatting, but I'll spare you that stuff.  However, the bottom half of my last email to Joe got into a discussion of sports modeling and betting lines, and since Joe didn't know about that stuff, I should maybe reprint what I wrote to him here by way of prepping you for the series itself.  I've started using a lot of betting terminology in my write-ups, but it was maybe a mistake to assume that everyone knows what the Hell all of that means.

By the way, I think we have a pretty good project.  I am totally excited about it.

FPI is measured in points against a nominally "average" team played on a neutral site.  So if the Giants are +3, then that means they would be a field goal favorite against an "average" NFL team if the two teams played at my local high school.  These numbers are based on results from a mathematical model developed by some stat geek at ESPN, which they then plug into each team's schedule to determine predicted wins and losses.  Football Outsiders also has a model (S&P+), and the bookies at Vegas have one, too.

Vegas uses these numbers to set betting lines.  If the Giants are +3 and the Redskins are -1, and they play at Giants' Stadium (+3 for home field), the Giants become 7 point favorites.  The line becomes Giants (-7) or Redskins (+7).

The vigorish (vig) is the cost of gambling.  You typically have to lay $1.15 or $1.20 to win $1.00 because the house charges for running the book.  But they also use the vig to shade lines when they aren't confident enough to move the line a whole half-point.  For example, the Bills are at 6 wins, but the vig is -130/even, which means that, really, the numbers predict the Bills will win something like 6.25 games.  That's meaningless in terms of betting odds, but the books aren't going to just give you .25 games of mathematical wins-expected, so they moved the vig to make it slightly more expensive to take the Over.  If you take Over 6 games, you have to bet $1.30 to win $1.  If you take the Under, you bet $1 to win $1.

Speaking personally, I think ESPN's FPI is mostly useful from the standpoint of understanding expectations.  FPI is extremely responsive to recent results, which I tend to like.  Both Football Outsiders and SB*Nation use S&P+, which leads to endless explaining after the fact why Bad Team X beat Good Team Y last Sunday despite the fact that this isn't what the model predicted.  I HATE that.  I mean, I really, really hate it.  Finally, the guys in Vegas are really only trying to be generally close on most games, so that they can balance the action.  They are really, really good at that, and they never apologize after the fact, but then, this is a living for them where it's basically just a hobby for the other guys.

I don't personally bet on sports, but I got interested in it through Ross Tucker's EVEN MONEY podcast because I find the math to be really interesting.  I feel like I've learned a lot about sports from listening to Ross's in-house pro, Steve Fezik, talk about how he sets lines, and that has influenced the way I write about this stuff.

Long answer to a short question.  Sorry about that.


  1. Great I know. These indexes help quantify thoughts you develop intuitively while following the sport.

    1. Ha! Glad to help.

      Like I said, I love the math behind this stuff. But I also think guys get jammed up when they try to explain why Team A "shouldn't have won" because the model says they weren't the better team. You see that occasionally from the sabermetrics crowd.