How much do we tend to know after five weeks? Bill Connelly compares five-week data to full-season data to find out if we should be worried about TCU and Baylor.
04 Nov 2004
by Aaron Schatz
Following a 42-14 shellacking of the Oakland Raiders on Sunday, the San Diego Chargers stand at 5-3 and are tied for the lead in the AFC West. They already have won more games than they did all of last season, and more games than the oddsmakers predicted in their preseason betting lines. Among the major surprises of 2004, this ranks somewhere between the Red Sox coming back from down three games to none and the resignation of Jim McGreevey.
Part of the surprise is that the Chargers are winning with the pass more than with their outstanding young running back LaDainian Tomlinson. Quarterback Drew Brees is having a phenomenal season, with his TD-to-INT ratio of 14-to-3 second only to Peyton Manning and a passer rating of 106.8 that ranks him only behind Manning and Daunte Culpepper. (You'll notice he ranks eighth according to our Football Outsiders ratings, in part due to an easier schedule of opposing defenses.)
With Brees on pace to throw for 3194 yards, the Chargers have to be wondering if they made a massive mistake on draft day by drafting quarterback Eli Manning and flipping him to the Giants for another rookie, Philip Rivers, and a selection of additional picks. How would the Chargers look today with rookie phenom Roy Williams added to their receiver corps, or safety Sean Taylor roaming the secondary, or tackle Robert Gallery helping to protect Brees and make holes for Tomlinson? (Admittedly, drafting Gallery would have meant not trading down to accumulate additional picks from the Giants.)
The Chargers could have used any of these players to improve other aspects of the team if they had expected this kind of turnaround from Brees. But was there any reason to expect such a thing?
Well, maybe not a season quite this good, but there were in fact a number of indicators that said that Brees and the Chargers offense were going to be much better this season, starting with last year's poor performance on third downs.
In 2003, the Chargers' passing game was one of the worst in the NFL when it came to converting third downs. (This was addressed in our preseason DVOA projections.) Even a small improvement in that department can become a major improvement in total offense. Suddenly, a few drives that ended early get a chance to go longer, and "longer" could mean anything from a few extra yards to a touchdown. Therefore, as a general rule, offenses that are among the league's worst in converting third downs tend to improve the next season, while offenses that are among the league's best tend to decline.
Brees converted only 23 of 102 third-down passes last season for first downs or touchdowns. This season he's 27 out of 58, including 15 to the NFL's hottest young tight end, Antonio Gates. On the whole, both rushing and passing, the Chargers have gone from converting 31.5 percent of third downs to 43.0 percent of third downs.
A look back at quarterbacks of the past might also have hinted that the Chargers were giving up on Brees a bit too early. As a general rule, young quarterbacks who are not named "Roethlisberger" do not play extremely well in their first few seasons, but improve each successive year. That was not the case with Brees, who dropped from a mediocre passer rating of 76.9 in 2002 to a miserable rating of 67.5 in last season. (In our ratings for Football Outsiders, the drop was equally poor, from 22.0 points above replacement level in 2002 to 14.7 points below replacement level in 2003.) This kind of regression by a young quarterback is uncommon, but not unheard of, and a look at NFL history should have clued the Chargers in to the idea that Brees would bounce back.
This is where we get to introduce a statistical analysis method that is new, and at the same time as familiar as an old sweater or the arrest of a Portland Trailblazer: similarity scores. (Click here if you want to skip the explanation and go straight to the analysis.)
Similarity scores were first introduced by Bill James to compare baseball players to other baseball players from the past. The general idea was to start at 1000 points and subtract for the various differences between two players; the players closest to 1000 were the most similar. The method is all over the great Baseball Reference website and, just as UNIVAC eventually led to your Palm Pilot, can be seen as the ancient predecessor to advanced baseball projection methods like Nate Silver's PECOTA.
It was only natural that the idea would spread to other sports as statistical analysis spread to other sports. John Hollinger introduced a version last year in Pro Basketball Prospectus and you'll find an advanced version in the new Pro Basketball Forecast 2004 (which, by the way, you must purchase if you enjoy the NBA; Hollinger is both an innovative analyst and a witty essayist). And in football, completely independent of each other, three different analysts created their own version of similarity scores during this offseason.
The Football Project guys, our co-conspirators on Pro Football Forecast 2004, use their method to predict the future of Steve McNair's career in the Tennessee chapter of PFF (a prediction for decline which looks pretty on target at this point). Doug Drinen, the master of the Pro Football Reference website, developed a set of similarity scores for the fantasy website Cheatsheets.net. And I developed a whole set of similarity scores, originally to figure out what would happen to Terrell Owens upon being traded to Philadelphia, and eventually expanded to TE, RB, and QB.
I planned on writing long, detailed articles with a set of similarity analyses for each position, but like a lot of my plans for long, detailed articles I ran out of time before the season started. I was just about to start writing the first one when Drinen published his similarity score material, which made me feel a little weird, and is part of the reason the whole thing got pushed off -- although Drinen's method is much different than mine or FP's because he concentrates on fantasy performance, including the use of "Value Based Drafting" value to determine similar players. The method showed up quietly in a couple of articles, such as the piece on the decline of wide receivers drafted in 1996 (who, with the exception of Owens, have continued to decline).
So we introduce it here, on a limited basis. By limited, I mean please don't ask me to run lots and lots of these for the next few weeks -- I have way too many other things to do during the season. The system may show up here and there for the next three months, but consider this an appetizer, and we'll do a lot more with this next preseason.
A few other caveats: This method compares standard statistics like yards and attempts, which are of course subject to all kinds of biases from strength of schedule to quality of receiver corps. The database for player comparison begins in 1978, the year the 16-game season began and passing rules were liberalized (you'll find I use this a lot as a starting point for the "modern" NFL). I've attempted to project statistics for 1982 and 1987 as if the strike didn't happen (which is hard due to injuries and certain players who crossed the 1987 picket line to play more than 12 games). The method is subject to change in the future. Here is how it works right now:
Now that we've explained the similarity scores system we're using, we can go back and use that system to compare Drew Brees to other quarterbacks who have followed similar career paths. There are a number of quarterbacks who, like Brees, followed a league-average performance in their first full seasons as the starter with a step backwards instead of a step forwards. As I noted earlier, a look at NFL history should have clued the Chargers in to the idea that Brees would bounce back. Frankly, a look at Chargers history should have clued the team into the idea that Brees would bounce back, because among the most similar players to Brees is a name of particular interest to San Diego fans: Stan Humphries.
In his first season as the starter in San Diego, 1992, Humphries completed 58 percent of his passes for 3356 yards and 16 touchdowns. Like Brees, he took a step backward the next year, completing 53 percent of his passes for 1981 yards and only 12 touchdowns. Like Brees, he had to share time with another quarterback after his regression (Doug Flutie for Brees, John Friesz for Humphries).
And then, in 1994, installed as the starter from the beginning of the year, Humphries led the Chargers to their only Super Bowl appearance. His numbers were nowhere near as good as what Brees is doing right now, but they were back to the level of 1992.
Another quarterback with a career path similar to Brees prior to this season was Washington's Mark Rypien. Rypien's first full-time season as the starter with the Redskins was 1989, during which he completed 59% of his passes for 3768 yards. But he struggled with injuries in 1990, playing only 11 games and completing only 55% of his passes for 2070 yards (coincidentally, the second-year backup that took over when he was injured was Stan Humphries).
In 1991, back as the starter for Washington, Rypien rebounded to complete 59% of his passes for 3564 yards and a career-high 8.5 yards per pass. That year, the Redskins went 14-2, won the Super Bowl, and are now considered one of the greatest teams of all time. Remarkably, further down the list of similar quarterbacks is a third quarterback who went to the Super Bowl in his third season after a two-year stretch similar to Brees' 2002-2003, Neil O'Donnell of the 1995 Pittburgh Steelers.
Now, not every quarterback who had a similar two-year stretch made the Super Bowl in the third season. The list of similar quarterbacks (found at the bottom of this article) includes a number of players who had long careers where they never quite reached Pro Bowl status: Steve DeBerg, Jeff George, Jake Plummer. But with one exceptions, Tony Eason and Randy Wright, each of the quarterbacks on the list rebounded to have a fairly long and at least marginally successful career (and the Wright similarity is a little kooky because it involves the strike year).
The similarity scores can also be used to analyze just how good Brees has been this season. Double Brees' numbers for 2004 to get a 16-game season and one season stands out as by far the most similar. Chad Pennington was a year younger when he had his breakout season in 2002, and a couple of his games came in relief of Vinny Testaverde rather than starting, but his numbers are quite similar to those being put up by Brees. Pennington, of course, led the Jets to the AFC East title two years ago and is considered one of the best young quarterbacks in the game. The list of similar seasons also includes three Troy Aikman years and, for you pessimists, two more Tony Eason seasons.
The Chargers can enjoy their run this season, and perhaps an unexpected trip to the playoffs, but long term they have backed themselves into the same corner as the Cincinnati Bengals of 2003. They have a multi-million dollar young quarterback sitting on the bench while a veteran has put them back in contention with a performance that matches the league's top quarterbacks. Unlike the Bengals, however, the Chargers won't get to bench Brees to give Philip Rivers his chance. Brees is a free agent and will be the number one commodity on the market next season. His numbers say that he will be a quality starter for whichever team offers him the most attractive contract. And with Philip Rivers, the Chargers will return to square one.
|Quarterbacks similar to Drew Brees in 2004|
|CAR = year of career SIM=similarity score, closer to 1000 being more similar
(Note: Brees projected over 16 games)
|Quarterbacks with similar two-year patterns to Drew Brees in 2002-3,
along with their numbers for the season following
|CAR = year of career SIM=similarity score, closer to 1000 being more similar
First season listed is compared to Brees 2002, second season to Brees 2003
(Note: Brees 2004 statistics projected over 16 games)
This is an expanded version of an article that originally appeared in Wednesday's New York Sun.
1 comment, Last at 21 Apr 2006, 2:06pm by Software