## NFL Week One Recap

While I didn’t make a lot of predictions for week one, I did use a model to predict if the home team will win their week one game based on their point differential from the season before.  In the future I’ll have win, point differential, and total point predictions.  Going forward I’ll also be using another person’s predictions as a comparison: Bill Simmons.  Bill has apparently been betting on the NFL for 20 years now and has a nation-wide reading audience, so you might think that he knows what he’s doing (unless you read his columns, in which case you know he doesn’t).  Bill only picks against the spread, and he has to do it on Thursday, which I would not recommend.  But to be fair in the future, I’ll look at my performance against Bill using his spreads, and then how I do against the Bodog spreads and other lines.

First, let’s pick some winners.  The ROC for my model has the best accuracy when you use a cut-off at .43.  So as long as the home team is predicted to win 43% of the time or more, take the home team.  If I do that, here’s how my model did in week one:

That’s 11-5, which is pretty much what you’d expect from how the model did in the past (about 66% accuracy).  This isn’t really how you should evaluate a probabilistic model; you’d want to make a calibration curve that shows that teams win as often as the model says they should.  So, for example, if you take all the times that the model says a team will win 40% of the time and get their win percentage, it should be close to 40%.  I can’t really do this with 16 games, since there aren’t enough data points to work it out.  But, if you just want to win your NFL pick ’em league, the binary outcome is good enough.

In my previous post, I said you shouldn’t use the winners to pick against the spread because the accuracy isn’t high enough.  But just so we can take a look, let’s assume you did.  Let’s see how it went, based on Bill’s picks.

Saints (-5) over Vikes (I agree): push

Giants (-7) over Panthers (I agree): winner

Packers (-3) over Eagles (I disagree): winner for Bill, loser for me.  The model doesn’t know that I think letting McNabb go was a horrible mistake, although if Vick gets to play a lot and is about as good as he used to be, the Eagles might turn out ok.

Dolphins (-3) over Bills (I disagree): winner for Bill, loser for me.  Even statistics can’t account for the Bills.

Falcons (-2) over Steelers (I disagree): loser for Bill, winner for me.

Lions (+6) over Bears: (I disagree): winner for Bill, loser for me.  The Lions almost won outright in addition to covering, except for a technically correct yet horrible rule.  Is it just me, or didn’t it used to be enough to catch the ball in the endzone and get two feet down?  Or even two feet and a butt?  How about two feet, a butt, and rolling over a bit?

Bengals (+4.5) over Patriots (I disagree): loser for Bill, winner for me.

Colts (-2) over Texans (I disagree): loser for Bill, winner for me.

Raiders (+6) over Titans (I disagree): loser for Bill, winner for me.

Jaguars (-2.5) over Broncos (I agree): winner

Browns (+2.5) over Bucs (I agree): loser.

49ers (-3) over Seahawks (I disagree): loser for Bill, winner for me.  Bill isn’t a big fan of Pete Carroll, but it doesn’t matter much when the other team can’t run, throws interceptions, and commits 8 penalties.

Cardinals (-4) over Rams (I agree): push.  If you happened to not watch this game, it was as ugly as I thought it would be.  Which was very.

Redskins (+3.5) over Cowboys (I agree): winner.

Jets (-2.5) over Ravens (I agree): loser.  I’m honestly sort of happy about this one.  I’m not a Ravens fan, or against the Jets per se, but I’m definitely on Ray Lewis’ side in terms of the “let’s see if they can win anything before we go too crazy” camp.  They only went 9-7 last year, guys.

Chiefs (+5) over Chargers (I disagree): winner for Bill, loser for me.

So overall Bill went 7-7 with two pushes, and I went 8-6 with two pushes.  If you got each game at -110 and bet \$110 on each, following Bill’s advice would have cost you \$70 while following mine would have won you \$140.  So I came out slightly ahead this week, and one game ahead of Bill.  More exciting predictions next week, and in the meantime some stuff on the details of bet size and other things that go into (or could come from) my algorithm.

This entry was posted in Uncategorized and tagged , , , . Bookmark the permalink.

### 3 Responses to NFL Week One Recap

1. nerdnumbers says:

Alex,
Awesome stuff! Nice to see you outperform someone more listened. Also its cool given you didn’t go the “Deep Blue Conspiracy” route and use your human knowledge to override a computer decision (McNabb). Happy to see a blog using R and ROC curves (the backbones of my masters thesis), I just wish I was a bigger football fan. Even your model realized Denver was a bad choice though, didn’t you tell it about Tebow?

• Alex says:

Thanks Andres! It’s really tempting to go against the model sometimes, but the way I think of it is, why have a model if that’s what I’m going to do anyway? I don’t put in injuries or trades or anything, so the model averages over that and presumably ‘accounts’ for it somehow. And it’s always possible that Kolb would have been better. So I try to always follow what my models say. Which is also why I stuck with Jacksonville against Denver, even though I definitely would’ve gone the other way (not because of Tebow, though). There should be more ROCs in the future!

2. Pingback: NFL Week 1 Recap | Sport Skeptic