The Two Point Rule


Version dated December 16, 2007, by David W. Myers.

Using data obtained from NFL.com, we have looked at season ending statistics from 2001 to 2006 and I noted that using median point spreads and win records was a more accurate predictor of who would win in the post season than win records and average point spreads. Medians alone in this period are more accurate, but the most accurate predictor so far is the use of median point spreads and the two point rule.

The two point rule works as follows:

2002 Season

So for now, we'll take a look at the ending records of the 2002 NFL season.

RankTeamMedianSpreadWinsLossesTies
1 TB 13.5 9.38 12 4 0
2 PHI 11.5 10.87 12 4 0
3 GB 6.0 4.38 12 4 0
4 OAK 12.5 9.13 11 5 0
5 TEN 6.5 2.69 11 5 0

In 2002, the AFC winner was Oakland, and the NFC Winner was Tampa Bay. Using averages and records, you would predict Philadelphia to be the winner in the NFC that year. But while Philadelphia has the same record as Tampa Bay and a better average point spread, it is lower than Tampa Bay in median point spread. Tampa won the NFC championship and the Super Bowl. The two point rule predicts both AFC and NFC winners and the Super Bowl winner.

2004 Season

In 2004, the final season rules are as follows:

RankTeamMedianSpreadWinsLossesTies
1 PIT 9.5 7.56 15 1 0
2 NE 12.5 11.06 14 2 0
3 PHI 10.5 7.88 13 3 0
4 IND 9.5 10.69 12 4 0
5 SD 7.0 8.31 12 4 0

In 2004, Pittsburgh has a 15-1 record, but New England has a 14-2 record. New England's median point spread is 12.5 points per game, while Pittsburgh's is 9.5. The rule predicts that New England is superior and that it should win the AFC and defeat Philadelphia in the Super Bowl.

These are the two best results of this system. I only have data for 2001 through 2006 so at this point we'll discuss a logical issue with the rule and then systematically go through all the seasons in turn and evaluate the results in terms of the two point rule.

Caveats

One issue with the two point rule is the generation of logic loops. This makes it impossible to write programs to calculate the two point rule directly. We'll give an example of a two point rule loop.

RankTeamMedianSpreadWinsLossesTies
1 A 7.0 7.00 15 1 0
2 B 10.0 11.00 14 2 0
3 C 13.0 9.50 13 3 0

In this example, Team B should be able to beat Team A. Team C should be able to beat Team B. But Team A, by virtue of the two game lead it has over team C, should be able to beat Team C. In this circumstance, it is impossible to predict a winner without knowing who plays whom first.

2001 season

RankTeamMedianSpreadWinsLossesTies
1 STL 13.0 14.38 14 2 0
2 PIT 8.0 8.75 13 3 0
3 CHI 6.5 8.44 13 3 0
4 GB 8.0 7.75 12 4 0
5 SF 4.5 7.94 12 4 0
6 PHI 7.0 8.44 11 5 0
7 MIA 5.5 3.38 11 5 0
8 NE 5.0 6.19 11 5 0
9 OAK 4.0 4.50 10 6 0
10 BAL 3.0 2.38 10 6 0

The 2002 season has the biggest upset of all the years. New England starts ranked 8th. Saint Louis is predicted as the NFC winner and Super Bowl champion. Pittsburgh is predicted as the AFC champion. In between Pittsburgh and New England are a bunch of NFC teams that Saint Louis defeats. Oakland, in this scenario, isn't considered as good a team as New England and probably should have lost to New England in the playoffs. New England wins with two consecutive upsets to win the Super Bowl.

2003 season

RankTeamMedianSpreadWinsLossesTies
1 NE 6.5 6.88 14 2 0
2 KC 7.0 9.50 13 3 0
3 TEN 7.0 6.94 12 4 0
4 PHI 7.0 5.44 12 4 0
5 STL 5.5 7.44 12 4 0
6 IND 3.0 6.94 12 4 0
7 CAR 3.0 1.31 11 5 0
8 GB 8.5 8.44 10 6 0
9 BAL 5.0 6.88 10 6 0
10 SEA 4.0 4.81 10 6 0

In this season the two point rule predicts the AFC champion and the eventual Super Bowl champion but does not predict the Cinderella season Carolina has in the playoffs. After defeating Dallas (ranked 12th) the Panthers go on to upset the Rams and the Eagles to get to the Super Bowl.

2005 season

RankTeamMedianSpreadWinsLossesTies
1 IND 15.5 12.00 14 2 0
2 DEN 9.5 8.56 13 3 0
3 SEA 5.0 11.31 13 3 0
4 JAC 6.0 5.75 12 4 0
5 PIT 11.0 8.19 11 5 0
6 NYG 8.0 6.75 11 5 0

In this season the two point rule predicts that Seattle will win the NFC and that Indianapolis wins the NFC and the Super Bowl. Instead, Pittsburgh has their magical end of season and wins the AFC and the bowl game. One item of note is Pittsburgh's exceptionally high median point spread. The New York Giants, which have a higher median point spread than Seattle, are predicted to lose because they can't overcome the 2 game difference in record.

2006 season

RankTeamMedianSpreadWinsLossesTies
1 SD 8.0 11.81 14 2 0
2 CHI 10.0 10.75 13 3 0
3 BAL 10.0 9.50 13 3 0
4 NE 7.0 9.25 12 4 0
5 IND 4.0 4.19 12 4 0

This is the season Indianapolis gets their revenge. In passing, it's worthwhile to note that Indianapolis recovers their run stopping safety Bob Sanders just before the playoffs and their whole play seemed elevated into the playoff run. No system I am aware of really takes injuries into account, and that would have been necessary to predict an Indianapolis run.

In terms of the system, there is no difference between the top three teams. They are rated evenly. Indianapolis upset 3 of the top 5 teams, with New England upsetting San Diego. But in the context of the two point rule, Indianapolis's defeat of Baltimore is as impressive as their defeat of Chicago.

Comparisons

In the table below, we show the differences in predictive power of three possible methods. The first is record and then point spread, the second is just median point spread (ignoring records), and the third is the two point rule. Again, I only have the 6 years of data, but in those six years, the two point rule has a small but real advantage.

AveragesMedians aloneTwo Point Rule
AFC Champion 223
NFC Champion 445
Super Bowl Champion 123

Now, obviously, the two point rule isn't perfect. But it seems to offer, in this sample set, a more accurate way to identify potential winners than does elementary wins plus point spread analysis.