## Friday, February 4, 2011

### Packers and Steelers fans, enjoy this Sunday

-- because "We'll be back next year" isn't so likely any more.
by Jim Glass

Once upon a time, way back before 1998, a team playing in the Super Bowl could reasonably expect to be in the thick of the competition for the Lombardi Trophy again the following year.

From 1970 through 1997 the Super Bowl team conference champions won an average 79% of their regular season games, and during the following year won 69%. In the follow-up seasons they compiled 47 winning records against only six losing and three .500 records. As 69% is 11 wins in a 16-game season, the teams could feel confident that they'd be in the playoffs again.

Then in 1999 the defending Super Bowl champion Broncos and runner-up Falcons, after going a combined 28-4 during 1998, fell to 6-10 and 5-11 respectively.

From then on, defending conference champions not named "Patriots" have had only 10 winning records in 20 seasons, against nine losing and one .500 record. All 24 Super Bowl participants including the Patriots have compiled a following-year average 56% winning record, a tad below 9-7.

 SB Teams W - L% Next year season record

 SB year Next year winning 0.5 losing 1970-97 0.786 0.692 47 3 6 1998-09 0.784 0.560 14 1 9

Is this just a fluke among the small number of Super Bowl teams or a league-wide pattern? To see, I took a simple measure of regression to the mean for playoff-quality teams -- all teams that won 11 or more games in a 16-game season or 10 or more in a 14-game season -- and applied it to all of them back to 1970.

The measure is just the percentage by which a team's record retreats towards .500 from one year to the next. For instance, if a team is four wins over .500 in the first year and the next year it wins two fewer games it regresses by 50%; with five fewer wins it regresses by 125%, if it wins one more game it regresses by -25%. Call this measure "win regression".

Over the 12 years 1998 to 2009, teams that won 11 or more wins regressed by an average 75%. (The dominant teams that won 14 or more games regressed the same 75% as the lesser-winning playoff teams). And the trend has been up -- over the last three seasons the figure is 83%. That's a lot. With win regression of 83%, the next 16-0 team will be expected to go 9-7 the year after.

It seems that 1998 was a swing year of some sort. Because one-year samples of 11+ win teams are so small as to be highly erratic, I took five-season moving averages of win regression (the year plus two before and two after). For 1997, by this measure, win regression was 49%. The next year it jumped to 62%, and it's risen steadily since to the current 75% (or 83%), an increase of 50% (or 66%).

What's behind this? My first thought was the obvious one: free agency and the salary cap. With many more players moving, and teams having more equal resources, good teams logically would seem to have a harder time staying on top.

Bill Walsh's 49ers famously had both Joe Montana and Steve Young at QB for six full seasons. When Montana got hurt Young went in. But no team can afford two such players at the same position today, and players that good won't stay to sit. When Tom Brady got hurt after the 16-0 season the Patriots had to send in a backup who hadn't started a game since high school (and when he played OK they had him bid away). If instead they'd had another Hall of Famer to send in as Tom's replacement, their "regress" the following year might have been a lot less than to 11-5. So it seems that free agency and the salary cap are a very logical explanation for a big increase in win regression.

But they came into effect in 1993, five years earlier -- and that was at a prior peak of win regression. From 1989 through 1992 win regression ran at 70% to 73% -- and in 1993 it started falling. By 1997 it was down to 49%, a decline of near a third. That doesn't fit the "free agency and salary cap" explanation at all.

What else to say? In the 1970s win regression ran at only 30% -- it was a lot easier to maintain a dynasty in those days. After that, a picture probably describes better than words.

The vertical line on the chart is 1992, the last year before free agency and the salary cap.

The overall trend is clear, one can draw a straight line from 30% in 1972 to 80% in 2010. But the ups and downs may seem odd because they don't at all fit the story of "parity", as evidenced by win regression, being caused by the salary cap and free agency. The steady rise in win regression started a good two decades before then. When free agency and the salary cap did arrive, win regression plunged for half a decade. After which, for no obvious reason, it started heading back up again to higher than ever.

What explains these turns in the curve? I have no idea, except perhaps simple randomness in annual numbers around the trend line. (The chart doesn't look unlike a stock market chart.) The number of 11+ win teams each year is quite a small sample.

What's been the force behind the rising trend line in win regression, if it hasn't been free agency and the salary cap? There are a lot of possibilities, you probably can come up with at least as many good ones as I can.

But the final result of it all is that we can expect win regression of 80% or more from this year to next. So the word for Packers and Steelers fans is: Carpe diem!

Jim Glass said...

For the record, I do have a possible explanation for the rise in win regression starting in the 1970s, including the counter-intuitive big drop with the arrival of free agency and the salary cap. I didn't put it in the post because I tend to blather on as it is. But FWIW...

There are serious quality differences in the managements of different teams, most will surely agree.

In the early days of the NFL these were very large, as owners sometimes hired drinking buddies as coaches, and often hired celebrity-name former players. With revenue coming from game-day ticket sales, the owners didn't have much incentive to do otherwise. There are histories that say the main reason why Lombardi and Landry dominated for a generation was simply that they were a lot smarter than the competition. If in the 1950s your team had Paul Brown, or in the 60s and 70s you had Lombardi/Landry/Shula, you were on top and going to stay there.

But when the big money started coming in during the 1970s, owners gained a lot more incentive to improve the quality of their investments. The worst teams improved their managements, which means they closed the gap with the best, so regression up from the bottom and down from the top both started steadily increasing.

Then free agency and the salary cap arrived in 1993. The notion that these should further increase parity across the league, equalizing teams, assumes that all team managements would be equally adept at using them. But if some teams were much better at it than others, the new rules could be a new club for the best-run teams to use to beat up on the poorly run teams, reasserting their dominance.

The big-name signings of the day seem to fit the pattern: Reggie White to the Ron Wolf-run Packers, Deion Sanders to the 49ers then Cowboys, etc. Arithmetic suggests these teams must also have been cutting their deadwood to make cap space. It's "marginal returns" again, the best managements first to fully understand the new rules and first into the FA pool getting the best players at the lowest price. These "early adapters" used FAs to cement their leads over the laggers, and win regression significantly decreased.

But by 1998, after five years, the laggards had caught on and all the teams were back on an even playing field as to FAs and the salary cap, so win regression started rising again. Today the salary cap and free agency probably are in fact driving increasin parity across the league, and the record-hight 80% win regression rate.

OK, as to the chart's big drop then rise after 1992, I admit all this is a mere just-so story made up after the fact -- but I like it better than a big random variation. Just-so stories are fun to make up and sometimes can even be true.

Free conference calls said...

Anonymous said...

I am just as baffled to explain the data but my sense is that we should try to move away from our reliance of the "Free Agency explanation" when we do this type of analysis. Mr. Glass may be on the right track but FA creeps into his analysis near the end as well. The point made about increased competitiveness between owners resulting in greater parity may be the simple answer. Increased parity and innovation in off/def schemes make it hard to prep for different opponents each week, therefore brigning about unexpected losses leading to win regression. That's all I got.

Brian said...

Could we just look at the year-to-year regression for all teams? If the top teams are losing more games the next year, then it would reasonable to assume that the bottom teams are winning more the next year. If you don't limit yourself to 11+ win teams, your sample size goes up quite a bit, and maybe the trend becomes more clear. Does that make sense?

Bigmouth said...

Another thought-provoking article by Jim Glass!

Jim Glass said...

Bigmouth, thanks for the kind word.

Brian, the teams at the bottom have been regressing up to the mean at a higher rate just as the ones at the top have been regressing down. But the method I used to illustrate what's been happening with the contender-quality teams doesn't work for all teams because the ones at the mean -- the 7, 8, 9 win teams -- move away from it. Their "anti-regression" has increased too, and the simple metric I used to measure regression gives bizarre numbers for them (and a "division by zero error" for 8-win teams). There are more sophisticated ways to deal with that ... but I was just interested in a clear picture of what's happening at the top with the contenders, and that's all my simple little method is good for.