The good news for our prediction system is that we came within ten seconds of going 11-2 in our predictions.   The bad news is, the ten seconds came in four different games.   We saw four games turn against us in the closing seconds on Sunday—well, three in the closing seconds and one in overtime—and San Diego’s convincing victory over New England in the day’s final game saved us from a sub-.500 day.  We were 7-6, making our prediction record, on the season, a very mediocre 24-17.  

            Will we get better?   Maybe; maybe not.   I’ll get into that.  It was a shocking Sunday. 

            San Diego 30   New England 10

                        We had it San Diego by 13, so. . .no big surprise there. (1-0)

            Green Bay 27  Seattle 17

                        We had it the Packers by 15. (2-0)

            Philadelphia 40  San Francisco 26

                        We had it the Eagles by 11.  (3-0)

            Dallas 24 Arizona 30 in overtime

                        We had it Dallas by one point.  (3-1)

            Jacksonville 24 Denver 17

A total shock.   We had Denver up 8.   I don’t know when the last time we got a Broncos game right was. . .I guess never.  We’ve been doing predictions for three weeks, and we’ve been wrong on the Denver game all three weeks.  (3-2)

            St. Louis 19 Washington 17

We had Washington as the third-best team in football, and St. Louis as the worst, thus Washington by 24.   Of course, everybody knows that coaching changes often give a team a short-term lift.  (3-3)

Indianapolis 31 Baltimore 3

Another total embarrassment for our rankings, which had it Baltimore by 6.  More in a moment (3-4)

            Tampa Bay 27  Carolina 3

                        We had it Tampa Bay by 1.  (4-4)

            Houston 29 Miami 28

Houston scored a touchdown with three seconds left in the game to put us on the wrong side of the line.   We had it Miami by 9.  (4-5)

            Jets 26, Bengals 14

                        We had it Jets by 8.  (5-5)

            Atlanta 22, Chicago 20

Chicago screwed up the last eleven seconds real bad; we had it Chicago by 13.  (5-6)

            New Orleans 34, Oakland 3

                        No problem here.   We had the Saints winning by 12.  (6-6)

            Minnesota 12, Detroit 10. 

                        Late field goal saved us on this one.   We had it Vikings by 17.  (7-6).

            In the rankings, this week was an earthquake.  Although many of the upsets were razor thin, our rankings have undergone sweeping changes as a result of Sunday’s games, much more than I would have anticipated this far into the season.   I don’t have much experience doing this, but my guess is that the ratings revolution that occurred on Sunday was rare, and won’t likely be repeated. 

            Let’s see.   A week ago we had Baltimore as the third-best team in the AFC, at 106.5, and Indianapolis two points under average at 97.8.   Now we have Indianapolis as the fourth-best team in the AFC at 104.5, and Baltimore under water at 99.7.   In other words, we now see Indianapolis as a significantly better team than Baltimore. 

            Now how in the world does that happen, in the middle of the season?    Indianapolis didn’t just win, they dominated, of course, but that’s not really what happened to the rankings.   Indianapolis also benefited from having played, in the two previous weeks, two other teams that also substantially exceeded expectations on Sunday.   Last week Indianapolis beat Houston 31-27, in Houston, in a memorable game in which Houston imploded in the middle of the fourth quarter.   But Houston upset Miami this week, which means that we now evaluate Houston as being better than we thought they were.  

            Two weeks ago, Indianapolis lost to Jacksonville, 21-23.   But Jacksonville on Sunday squashed the Broncos in Denver, which means that we now evaluate Jacksonville as being substantially better than we thought they were.    Houston and Jacksonville are up four points from where they were a week ago, and this pushes Indianapolis “schedule value” up as well.  It is the interaction of these things with the 31-3 Raven-mauling that projects the Colts so sharply forward.

            That kind of happened all around the league this week; the Week Six results interacted with one another to create tidal waves in the rankings.    One of the odder ranking results is that the Bears, despite losing at Atlanta in a game they were favored to win by 7, actually moved up in the rankings from fifth to second in the NFC.   Two of the teams ahead of them, Washington and Dallas, also lost, while Chicago’s ranking improved from 109.1 to 110.6. 

            Why?   Chicago’s previous schedule included Indianapolis, Tampa Bay, Philadelphia and Detroit.  All of those teams had good weeks, and all improved their rankings.   The collective gain in how we evaluate their opponents is larger than the underperformance of the Bears themselves.   It’s not illogical; it’s just counter-intuitive.

            These are our current rankings:

             We now see Tampa Bay as being the best team in football.   

            Monday night’s game doesn’t look close; we make it the Giants by 11.    But, as Al Michaels says every week, nobody knows nothin’ in the NFL, which certainly includes us. 

            The question that we can begin to study, perhaps within a few weeks, is “How reliably can one predict wins in the NFL?”   The answer, it would appear, is “Not very”.   We did these rankings last year for the NBA, and what I learned from doing that is that game outcomes in the NBA were much less predictable than I would have thought they were.  I suspect that the same may be true in the NFL. 

            Look at it this way:   Games would be predictable in the NFL if

            1)  We had perfect measurements of where every team was, and

            2)  Every team performed at its true level of ability every week.

            Of course, our rankings are not perfect and will not be, but we have to assume that, as we get more information in, they will become more perfect.   But we’re 24-17 in predictions now—below 60%.   We can expect to edge over 60%, I think, but I’m skeptical, at this point, about how far we’re going to get above 60%.  

            If you don’t know anything—if you’re just picking names out of a hat—you’re going to get half of them right.   How far forward you can move forward from 50% depends on three things.  

            1)  The error in your rankings,

            2)  The talent dispersal of the league, and

            3)  The degree to which each team tends to perform at a consistent level.

            In other words, if teams were typically separated by 20 points of ability and typically performed within a 3-point range of their true level of ability, then we could probably predict the games with 80% reliability.   But if teams were typically separated by 3 points of ability and had a normal up-and-down fluctuation of 20 points, then we probably would be lucky to get to 54%.

            Well, what are those numbers?  The standard deviation of team rankings is currently 8.1 points, meaning that the differences between competing teams tend to be about 8 points.   This figure is shrinking a little as the season wears on. When you only have a few games, you can sustain the illusion that your system is explaining all of the results, but as you get more results, it becomes more impossible to sustain the illusion that they all line up somehow.    This causes the standard deviation of the rankings to contract—8.7 points two weeks ago, 8.1 now. 

            Our current view of the most consistent and inconsistent teams in football is as follows:

            Most consistent:

            1.  Tennessee               2.13

            2.  Houston                  2.94

            3.  San Francisco          3.01

            4.  Philadelphia              3.19

            5.  Giants                      3.28

            Least consistent:

            1.  Jets                         9.71

            2.  Rams                       8.38

            3.  Kansas City              8.11

            4.  Denver                    7.82

            5.  New England           7.62

            These, of course, being the standard deviations of each team’s Game Output Scores.   (Incidentally, these scores could be used to study questions such as “are running-based teams more consistent than passing teams?” and “are strong defensive teams more consistent than strong offensive teams?”)

            Anyway, the standard deviations of performance consistency appear to converge on a point about 5.6, meaning that teams typically move up and down week to week by about 5.6 points.  

            In each game there are two teams, so what is relevant is the combined inconsistency of the two teams.  If each team has an “inconsistency score” of 5.6, then the two together would have an inconsistency score of 7.9.  (Take my word for it. . .I don’t have the patience to explain the math.)  

            That means that the difference between two randomly matched NFL teams averages about 8 points—while the performance inconsistency of the two teams is also about 8 points.   The “normal performance inconsistency” is about the same as the normal gap between the quality of the teams.   Therefore, it would seem, the games probably are not highly predictable.  

            If I was a better statistician, I could probably say that, given a certain standard deviation of teams and a certain level of consistency of performance, this is the level at which we should be able to predict the outcomes.  I’m not that good at formal statistical analysis; I don’t know how to work that one.   But just intuitively, I don’t think we’re going to be predicting 70% of the games accurately by season’s end.