Not a good week in the NFL predictions department; good week last year, not good the week before, poor week in Week Seven.   We were right on 8 games, wrong on 5, which is a normal percentage; we’re 25-14 on the season, 64%.   We’ll probably improve a little later in the season.   But it was a poor week because we lost most of the games on which we differed from Las Vegas.   Las Vegas had Houston over San Francisco, Pittsburgh over Minnesota, Cincinnati over the Bears, and Dallas over Atlanta, all of which was right.   We went the other way.   The only one we won was Buffalo at Carolina.   We also differed from Vegas on the over/under on five games, and again, we were wrong on four of them.   Bad week.

            I also had a glitch in my program last week.   I figure these rankings myself, in an Excel spreadsheet.   There was something in the spreadsheet that was supposed to be a formula that had been copied as a value, making the results a little off, given my own assumptions.   Not badly off. . .just a point or two.   Whether this contributed to the bad week, I don’t know.   It’s embarrassing anyway, but I never claimed to be an expert at this; I’m just trying to see what I can learn by doing the work.

            I’ll have predictions for this week later on, but first, a new wrinkle.    You know I’ve been predicting Points Scored this year by a new formula, which I haven’t actually explained yet.  But at some point it occurred to me, “Wait a minute, why am I doing that?   With the method I am using there, I could just as easily figure the team’s OFFENSIVE contribution to expected points, and their DEFENSIVE contribution.   Why haven’t I disaggregated them?”

            Damned if I know.   Sometimes I’m slow.   OK, so the process is this.   Each team has an offensive number (O Score) and a defensive number (D Score).   Ignoring the home field advantage for a moment, let us assume that Denver plays Baltimore this week, which they will or anyway are expected to.   Denver has an offensive number of 11.5, and Baltimore has a defensive number of 9.6   Adding those together, we have 21.1.  We thus expect Denver to score 21 points in the game.    Baltimore has an offensive number of 18.0, and Denver has a defensive number of negative 1.2.   Adding those together, that’s 16.8, or 17.   We expect Denver to win the game—ignoring the home field advantage—21 to 17.

            How are these numbers derived, you asked?   In fact, why don’t I invite you into the conversation on a more permanent basis, make this a dialogue between myself and. . ..Walt Whitman.  

            WHITMAN:  How are these numbers derived?

            JAMES:   Let’s say that each team starts with an offensive number of zero and a defensive number of zero.  It doesn’t matter what number you start with.   You’ll wind up in the same place no matter where you start.

            WHITMAN:   Like the rankings?

            JAMES:   Like the rankings.    But for purposes of illustration it is easier to start with some other number.   Let’s take last week’s Green Bay at Cleveland game.    Before that game, Cleveland came in with an offensive number of 3.7, and Green Bay came in with a defensive number of 8.1.   That’s 11.8, or 12 points.  

            WHITMAN:  Cleveland is expected to score 12 points.

            JAMES:  Right.   Green Bay comes in with an offensive number of 13.1, and Cleveland with a defensive number of 12.2.   That’s a total of 25.3.

            WHITMAN:  So Green Bay should win, 25-12. 

            JAMES:  Ignoring for the moment the home field advantage.   But Green Bay actually won 31-3.   Green Bay was expected to score 25.3 points; they actually scored 31.   What do we do with the difference?

            WHITMAN:  Split it between the teams?

            JAMES:   Exactly.    Their offense over-achieved, relative to previous expectations, by 5.7 points, actually 5.66 if you save another decimal.   Half of that—2.83 points—we attribute to Green Bay’s offense, and half to Cleveland’s defense.

            Green Bay’s offense has a “prior offensive number” of 13.1, and they over-achieved by 2.83 points.   That makes 15.9, actually 15.95.   Green Bay’s offensive number for the game is 15.95.

            Cleveland’s defensive number was 12.2, to which we also add 2.83.   Saving another decimal, it works out to 15.05, in this round of calculations’.

            WHITMAN:   Why is it different in the next round of calculations different?

            JAMES:   Because we have a different starting point for each team.   Sticking with one number, let’s use Green Bay’s offense.   Green Bay’s offensive number from last week was 13.12, which was based on the average of their previous five games.   Their “game output offensive numbers” from their previous five games totaled up to 65.62, which is an average of 13.12.   But when we add in the 15.95 from this game, that makes a total of 81.57, which is an average of 13.59.  So their “input offensive number” for the second round of calculations would be 13.59.

            WHITMAN:  Would be?

            JAMES:  It would be, but there’s another wrinkle.   At the same time that we’re re-calculating Green Bay’s offense, we’re also re-evaluating every other game played in the NFL this season. 

            WHITMAN:  Simultaneously?

            JAMES:  Well, not technically, but yes.   Technically, we pause the process after entering the scores from this week, and then re-calculate everything.  Then we repeat the process.

            WHITMAN:  So what does Green Bay’s Offensive number become, on the second round?

            JAMES:  Actually, it’s still 13.59.   But it could be something different, and it is something different, if you look at enough decimals. 

            WHITMAN:  Then what do you do?

            JAMES:   Then we repeat the process again, and again, and again, until the numbers entirely stop moving.    Eventually you reach a point where all of the output numbers are exactly the same as the input numbers, through several decimal points.   When we reach that point, we’re done. 

            WHITMAN:  Where do we wind up?

            JAMES:  Green Bay’s offensive number goes up, because they out-performed expectations, from 13.12 to 13.23.   They came in with a defensive number of 8.1, but allowed fewer points than expected; therefore their defensive number goes down to 6.83.    Cleveland has the opposite movements; their offensive number goes down to 3.33, and their defensive number goes up to 12.49. 

            WHITMAN:  So we’re done with that?

            JAMES:  Well, we should be done with that.   But as I was implementing this system, I didn’t know how to put in the home field advantage, so I had to research that.

            WHITMAN:  How did you do that?

            JAMES:   I started with a spreadsheet that had the scores of all NFL games this year, which actually isn’t a lot; it’s just 103 games.   I threw out the game played in London, between New England and Tampa Bay, which left 102 games. 

            WHITMAN:   And this leads?

            JAMES:  This leads to the realization that the home field advantage in the NFL this year has been tiny.    So far this year, in 102 games, the NFL home teams have outscored their opponents by only 139 points.   However, the better teams have been scheduled at home more often, slightly, which accounts for 34 of those points, so the actual home field advantage has been only 105 points.    Home teams have scored only 54 points more than they would have been expected to score on a neutral field, and have allowed only 51 fewer. 

            WHITMAN:  Shouldn’t that balance?

            JAMES:  We’ll get into that.

            WHITMAN:  Is the home field advantage really that small?

            JAMES:   Presumably not.   The home field advantage usually measures to some larger number, like 2.5 points per game, or 1.25 points per team per game.  

            WHITMAN:  So what do you do?

            JAMES:  I’m assuming that the larger number is the more accurate one, and I’m assuming—absent actual convincing evidence—that it expresses itself equally as an offensive and a defensive advantage. 

            Then we re-calculate.    We now have these figures:

            Green Bay is now expected to win, if they had a re-match at the same park next week, 25 to 11.   But since 25 to 11 is a bad football score, I’d probably make it 27 to 12. 

            WHITMAN:  How did you manage to put that chart inside a conversation?

            JAMES:  I’m very experienced at this.

            WHITMAN:  Then why didn’t you get the score right?

            JAMES:  Moving on.   With this new information—the home field advantage—we now have to re-calculate all games played so far this year.    

            WHITMAN:  Which leads to?

            JAMES:  More minor revisions in the O Scores and D Scores.   Minor, but not negligible.   Green Bay’s O Score drops from 13.23 (above) to 12.85.  

            By this method we can rate each team’s offensive and defensive performance on the season.   Of course, you can do that by just looking at how many points they have scored, but by doing it this way we can adjust for the quality of the defenses they have played against.  For example, Miami has scored 146 points this year in 6 games; Pittsburgh has scored 171 in 7 games, which is almost the same per-game average.   But Miami has faced far, far tougher defensive teams than Pittsburgh has.   Miami has faced only one team that is weaker-than-average defensively, and has faced four teams that are outstanding defensively.   Pittsburgh has faced teams that allow above-average numbers of points every week except one.    If you adjust for that, Miami’s offense is almost a touchdown a game better than Pittsburgh’s. 

            WHITMAN:  Who has the best offense in the league?

            JAMES:  C’mon; you don’t have to ask that.    Even guys who have been dead for a hundred years know that one.

            WHITMAN:   The Saints?

            JAMES:   Sure.   By far.   By our math, New Orleans’ offense is more than ten points a game better than any other in the league.  

WHITMAN:  And the best defense?

JAMES:   Denver.

WHITMAN:  Wait a minute.   You have Denver with a D Score of negative 1.2.

JAMES:  Yes.

WHITMAN:  And Washington with an O Score of a negative 1.9. 

JAMES:  Right.   You get a negative number if your offense scores or your defense allows less than one-half the normal number of points. 

WHITMAN:  What happens if Washington plays Denver?

JAMES:  Well, Denver played Oakland, actually.   We would predict a negative two points, but since that can’t happen, we just make it zero.

WHITMAN:   They actually scored. . .?

JAMES:  Three.   We would have predicted it 24 to nothing.   It was actually 23 to 3.

WHITMAN:   What’s the worst you have done this year?

JAMES:  Jacksonville at Seattle, October 11.   Knowing what we now know, we would have predicted that one at 24-17, Seattle.  The right number of points and the right winner, but Seattle won it 41-0. 

WHITMAN:  OK, is that all there is to know about the new system?  

JAMES:  Well, yes and no.   

WHITMAN:   There’s more.

JAMES:   There’s something I don’t understand.

WHITMAN:  Which is. .

JAMES:   In theory, each team’s rating should be equal to 100, plus their O Number, minus their D Number, right?   As best I understand the two systems—the old ranking system, which I have used for two years, and the new O and D number system—as best I understand them, they’re measuring the same thing.   If you’re five points stronger on offense than you are vulnerable on defense, you have to be five points above average, which means you should have a power rating in this system of 105.  

            And teams do—almost.    But there are small discrepancies for every team.    New Orleans has an O minus D separation of 20.02 points, but they rank at 119.6, not 120.02.  Every team has those small differences.   St. Louis is at 79.4 and 79.6, Denver is at 112.0 and 112.6, Tampa Bay is at 83.9 and 84.4, etc.   I don’t know where those discrepancies come from.

WHITMAN:   You don’t have a clue?

JAMES:  I have one clue.   First, the O Numbers in the NFL this year are slightly larger than the D numbers.    The O Numbers in the NFL average 11.01; the D Numbers average 10.68.   That’s not a problem.  It just means that, in the NFL this season, offense has had slightly greater control of the score than defense has, which is not unexpected and not terribly meaningful, at least until it is shown to persist over something more than a hundred games.

            But the difference between the O average and the D average, this week, is 0.3282 points, and the average discrepancy between the two rankings is 0.3282 points.    Last week the two discrepancies were both 0.35813 points.   

            Obviously, that’s not a coincidence.   They’re tied together.   The difference between an average offense and an average defense manifests itself, in this system, as a difference between the team’s rank, and their O Score minus their D score (plus 100).  I know that much.

            But why does it manifest itself in random amounts?  If it was 0.3282 for every team, maybe I could understand it, but it isn’t.  It’s 1.04 points for Miami—the largest in the league—and negative .68 points for Indianapolis.   It isn’t 0.3282 for any team; it merely averages exactly the same as the difference between the O scores and the D scores.   So far I’ve been unable to figure out why this happens.

            OK, let’s send Walt Whitman back to his Leaves-of-Grass-covered-Grave, and carry on with our report.  These are the updated rankings for the 32 NFL teams:

            New Orleans remains the league’s best team, in our opinion, and St. Louis has staked a clear claim as the worst.   Which is not easy, because I have seen Kansas City play, and they are terrible, just terrible.   I can’t imagine how bad a touchdown a game worse than that might be. 

            This is how I see this week’s games shaping up.  Once more, we see four games different than Las Vegas is betting them.  ..

1)  Denver at Baltimore

            Close game, could go either way; I have it 20-18 Denver.    Las Vegas is betting on Baltimore, about 23-19.    I’m kind of a Baltimore fan, a little bit, but I don’t see it.

2)  Houston at Buffalo

            Very close game.  I have it Buffalo 21, Houston 20.   Again, I’m splitting from Vegas, which is betting on Houston by 3.  

3)  Cleveland at Chicago

            Average/above average team at home vs. a bad team on the road, Bears should win handily.   I have 27-15, Chicago.

4)  Seattle at Dallas

            An upset wouldn’t shock us.   Seattle is up and down, very inconsistent.   Dallas 25, Seattle 19.   Betting line has Dallas as 9 and a half point favorites, which I don’t see. 

5)   Miami at the Jets

            Miami is actually a good team, but continues to play a killer schedule that may wipe them out of the playoffs by the mid-point of the season.   Jets 21, Dolphins 18.

6)  San Francisco at Indianapolis

            I think Indy’s not as good as people think, but I don’t think the 49ers are the team to beat them.   Colts 27, San Francisco 15.

7)  Giants at Philadelphia

            Dogfight.   I make it Giants, 27-26.   Here again, we’re parting from Vegas on both the winner and the over/under.   The bettors have it Philly, about 23-21. 

8)  St. Louis at Detroit

            Nobody can win an ugly-off with the Rams.   Lions 28, St. Louis 15.

9)  Oakland at San Diego

            Chargers 33, Raiders 13.

10)  Jacksonville at Tennessee

            Two of the worst defensive teams in the league, and two of the least consistent teams in the league.   At this time the Titans are the betting favorite, which is a bit of a surprise, in that Jacksonville is at .500 and Tennessee is at nothin’.   I agree that Jacksonville is not a good team and Tennessee could very well win it, but I have it Jacksonville, 27-23.    

11)   Favre returns to Green Bay.

            Green Bay 26, Minnesota 23.   Favre retires after the game.

12)  Carolina at Arizona.

            Blowout of the week.   29-9, Arizona.

13)  Atlanta at New Orleans.

            Somebody will beat New Orleans, but we probably won’t predict it.   Saints 37, Falcons 22.

            Other issues we usually look at here. . ..

            Bye weeks.   So far there have been 12 teams which have played after a bye week, and they have overachieved coming off the bye week by a total of 26.6 points, or a little more than 2 points each.   Still not enough data to mean anything much, but Seattle is coming off a bye week, and I believe they would probably beat the spread even without that, so. . ..

            Hot teams and cold teams. …temperatures are:

            There is still not much difference between “hot teams” and “good teams” at this point.    It’s more of a baseball concept, I think.

            Strength of Schedule. . ..we chop out the middle of the chart here, because it’s only interesting if you’ve played a tough schedule or a weak one.   Miami’s schedule has just been brutal.  They’d played almost the toughest schedule in the league last week, BEFORE they played New Orleans.  The league owes them 50 points:

            Comparing Miami to Washington, Washington has a 120-point advantage over the course of the season.   One-hundred-and-twenty-points.   Parity scheduling, my ass. . .

            Consistency. . . Not really sure we’re measuring anything meaningful here, but the league’s most consistent teams have been 1. Denver, 2. Tampa Bay, and 3. Minnesota.   The least consistent have been 1. Jacksonville, 2. Atlanta, and 3. Seattle.