Several years ago, Gregg Easterbrook pointed out that the record of the average pundit in picking NFL winners on a weekly basis is little better than chance; a simple glance at the year-to-date records of the pundits on any major television or radio network show (say, Fox Sports) will reveal that some of the pundits will be at or even below .500 on their picks for the season. In the effort to be clever, or ride the current trendy bandwagon, or just listen to their gut, many pundits are flat-out wrong more often than not. In fact, Easterbrook pointed out, in a recent year (I think it was 2009), the following simple algorithm posted a better predictive record than all but one published pundit:



Sounds absurdly simple, but it works surprisingly well. The algorithm tends not to be extremely effective for the first two weeks of the season, but it did go 10-6 in both Weeks 1 and 2 for a combined record of 20-12 (.625), which isn't bad. (By way of comparison, Peter King is 26-6 or .781 over that same span.) It will be interesting to see how effective the algorithm is as the season progresses.

However, I think there are algorithms that could predict weekly winners with even more accuracy, so I have devised a couple that I believe will be effective. These algorithms will test the trope that "Offense wins games, but defense wins championships." I am intrigued by the fact that teams with the better defensive passer rating have been overwhelmingly successful at winning championships over the past for decades. Last year alone, the team with the better defensive passer rating went 10-1 in the playoffs. Likewise, it seems as though the teams with offensive juggernauts are the ones who tend to dominate in the regular season; think Kurt Warner, Drew Brees, Peyton Manning, Tom Brady, Brett Favre. On the other hand, these teams don't always have the best records in the playoffs. Therefore, I predict that the offensive passer rating algorithm will have better predictive success in the regular season, while the defensive passer rating algorithm will be more effective in the playoffs. It will be interesting to see how it pans out.

Here are the two passer rating algorithms:





And here is an algorithm I personally think would have the highest success of all:



Feel free to contribute and track your own algorithms in this thread. I would love to compare the efficacy of a variety of predictive strategies. I would especially appreciate it if members would volunteer to track an algorithm so I wouldn't have to do all the work myself.

---

You are such an effing geek.

And this is coming from a guy who's last Facebook status was, "Its super embarassing when my wifey catches me pretending to be HeMan. I was raising my invisible sword and whispering, "By the power of GraySkull" and she caught me. :/ 

---

Easterbrook is showing off his econ background. It's a basic technique of prediction that economists, particularly those inclined to telling macro stories, love: use an average of past performance to predict future performance.

Query: Which, if any, of the following yield the same predictions as Easterbrook's algorithm?

1. Let Xi = 1 if a win in week i, 0 if a loss, i=1,...,n where Xn is the last game. Then Y (prediction for week n + 1) = "Win" if (∑Xi )/i> 1, "Lose" if (∑Xi)/i < 1. If ∑Xi/i = 1, pick home team.
2. Let Xi = 1 if a win in week i, 0 if a loss, i=1,...,n where Xn is the last game. If ∑Xi (team A) >∑Xi(team 😎, predict team A to win in week n+1. If ∑Xi(team A) =∑Xi(team 😎, pick home team.
3. Let X = wins through n games for team A, Y = wins through n games for team B. If X > Y, pick team A. If X
Signed,
Geezer (but senile) geek.


Edit: See, this is why the goddam "slapnuts" macro or whatever it is has to go.

---

And do not be conformed to this world, but be transformed by the renewing of your mind, that you may prove what is that good and acceptable and perfect will of God.
Romans 12:2 (NKJV)

Wade-it's possible to get around the macro. Check out the quote.

Looks like 2 & 3 would render the same results. Don't think 1 would as written, as it doesn't make any comoparison between the two teams.

---

Well, yes. The downside to this technique is that it is vulnerable to unexpected short-term fluctuations. It cannot compensate if, for example, Tom Brady or Aaron Rodgers were to go down for a game; it would predict the outcome of the game as though they were still in the game. The upside to the technique is that it is immune to the hot-hand fallacy so prevalent among sports commentators. For example, Tom Brady threw for 511 yards last week, leading a gushing Peter King to proclaim that he could walk right into the Hall of Fame now if he wanted, and no doubt leading other sports commentators to overestimate what he will do this week. Given that Tom Brady is only the sixth NFL quarterback in league history to throw for 400 yards in two consecutive games (the fifth, incidentally, was Cam Newton, so I have to think that last week marked the first time two quarterbacks had ever accomplished that feat on the same day), Tom Brady is much more likely to revert back to a number more closely approximating his career average of 246 yards per game than duplicate his career-best performance. This predictive technique refuses to be overawed by Brady's historic feat and stubbornly predicts that his performance from week to week will probably be typical instead of anomalous. It may therefore be inaccurate in the short term, but just as level-headed poker players usually take pokers chasing the hot hand to the cleaners, it will probably kick the asses of the bandwagon-chasing sports commentators over the long haul. (I remember being shocked to watch pregame shows last year and see that some TV commentators were sporting prediction records as low as 6-10 and 7-9; they would have been better off pulling names out of a hat or asking a chimpanzee's advice.)

That being said, here are the predictions for this week.

---

No sir, I do not see why the "goddamn" slapnuts must go.

---

---

Well, I guess we're a lock then to beat da Bears. 🖐

---

Fuker.

:)

---