View Full Version : The Steelers will win 9.6 games this season

07-30-2009, 11:05 PM
Blog and Gold

The Steelers will win 9.6 games this season

... That's according to the statistical gurus at Football Outsiders who, for going on six years, have meticulously charted essentially every single play run by every single team in every single game in the NFL. Their findings are fasciniating, especially for those of us who struggled to pass stats at Penn State.

Bill Barnwell, one of the site's authors advanced us the Steelers chapter of their 2009 Football Outsiders Almanac (on sale now!) and its an excellent read for any serious fan of football who understands that the game is as much an academic pursuit in its strategy as it is a physical one.

The Steelers chapter also gives an excellent historical summation as to how the Steelers have been able to maintain consistent success and excellence in the modern NFL and its probably the only time you'll hear the Steelers compared to a vintage Rothschild.

Some excerpts:

"The Steelers method is the "right" way to build a football team, in the same way that growing grapes on French hillsides and stomping them with bare feet is the right way to make wine ... The Steelers 2008 roster shows the hallmarks of a vintner's craftsmanship. Twenty-four of the Steelers top 30 regulars last year spent their entire career with the team ... Six of those players were first-round picks, but seven joined the team as undrafted free agents, and seven others were drafted in the 4th round or later ... The 2008 world champions were the greatest triumph of the Steelers Method ... The 2008 Steelers were deeply flawed but they won the Super Bowl because of their method ... Last year's Steelers were just another Steelers team: That's praise for the organization, not an insult to the reigning champs."

We were able to email Barnwell a handful of questions regarding Football Outsiders and its methodology, and how they arrived at 9.6 wins for the 2009 Steelers ...

BNG: The two statistics your publication frequently references are the DVOA & DYAR. What are they and how are they computed?

BB: The concept behind DVOA and DYAR -- our two most prominent statistics -- is actually pretty simple. For every play of the NFL season, we take the yards gained (or lost) and compare it to what the league average was for a similar player, after adjusting for down, distance, situation, and the quality of the opponent. We make further adjustments for bigger plays, scores, and turnovers.

The math behind it is relatively complex, but the concept's pretty simple; seven yards on 2nd-and-16 against the Lions when you're up 25 doesn't mean as much as those same seven yards on 3rd-and-6 against the Ravens in a close game.

We then split those numbers out two different ways. DVOA (Defense-adjusted Value Over Average) is a percentage that takes into account all the different plays an offense, a defense, a team, or a skill position player was involved in. The best offense in the league a year ago, by DVOA, was the San Diego Chargers. They had a 24.1% DVOA, meaning that they were 21.4% better than an average offense would have been in the same situations they faced. Last year, the Steelers had the best defense in the league, with a -26.9% DVOA. (The percentage is below zero because we're measuring defense; it means that the Steelers allowed 26.9% fewer yards and points in the same situations than the average team would have.)

DYAR (Defense-adjusted Yards Over Replacement) is a cumulative statistic, designed to measure a player's success versus that of a replacement-level player, by measuring the difference between the yards gained on the plays he was involved in versus the yards a replacement-level player would gain in that situation. (We define a replacement-level player to be someone playing at the level of a street free agent -- someone like Najeh Davenport, as an example.) A player who catches a touchdown pass on his only pass of the season might put up a ridiculous DVOA of 800% or something, but as good as his play was, it's only one play. DYAR is designed to reward players who repeatedly perform better than the league average. Drew Brees led the league with 1921 DYAR last year; in other words, given the 647 attempts Brees had last year (including sacks and aborted snaps, but not Hail Mary passes), Brees gained 1,921 more yards than a replacement-level quarterback would have.

For those of you who are baseball fans, think of DVOA as our batting average (rate statistic) and DYAR as our RBI (cumulative statistic).

BNG: You say the Steelers mean win projection is 9.6 games based on 10,000 simulations -- how do you run the simulations?

BB: We start with a regression analysis that takes into account hundreds of variables regarding each of the 32 NFL teams. The variables we use are ones that we've found to be true or relevant in the past; take third down performance, for example. We know that third down performance has a disproportionately high (relative to the other downs) impact to a team's success; that's obvious to anyone who watches the game. What we found, though, are that teams that are significantly better or worse on third down than they are on the first two downs tend to even out in subsequent seasons; a team that has a 0% DVOA on first down and second down but a 25% DVOA on third down, as an example, is likely to put up a third down DVOA way closer to 0% than they are 25% in the following season. Our regression analysis would note this, and adjust their projection of a team's offense down accordingly.

That produces an expected offensive, defensive, and special teams DVOA for each team. We use those figures (along with other variables, like home field advantage) as part of a simulation of each game of the NFL season; a team with a 30% DVOA might beat one with a -12% DVOA, say, 85 percent of the time. We simulate each game 10,000 times to account for the variance inherent to football (think about Santonio Holmes' feet being three inches off in the Super Bowl) and then that produces an average win total for each team.

Now, some teams have more variance in their expected wins than others. Pittsburgh, for example, has a 33% chance of winning 9-10 games, but a 37% chance of winning 11 or more games. (They have a 1% chance of winning 0-3 games.) Compare that to the Buffalo Bills, for example; we expect them to have a 55% chance of winning 4-6 games.

Of course, those 1% shots can come through; look at the Dolphins last year, who rode a perfect storm of their own health (Chad Pennington and Ronnie Brown making it through 32 games) and key injuries to their opponents (Tom Brady, for one) en route to an AFC East title. 1-in-100 is still a chance.

BNG: You state that Troy Polamalu is the best strong safety in the league by virtue of leading his position in 'defeats.' How do you define a defeat?

BB: A defeat is one of our individual defensive statistics, measured by the volunteers who work as part of our Game Charting project, which I'll get to momentarily. A player is credited with a defeat when he makes a play that either stops the offense from gaining a first down on third or fourth down, stops the offense behind the line of scrimmage, forces a fumble (regardless of who recovers the fumble), or intercepts a pass.

07-30-2009, 11:06 PM
Last year, the Ravens' Terrell Suggs led the league with 36 defeats. James Harrison was tied for fifth with 29, while James Farrior, LaMarr Woodley, and Troy Polamalu were all among the top 20.

BNG: How do you go about charting what each team does?

BB: That's our Game Charting project. Each week, we assign each of our flotilla of volunteers one half of a game; they tape the game, and then we provide them with the play-by-play. Using detailed instructions of what to look for and how to mark down particular plays, they note dozens of items and variables not included in the NFL play-by-play. That includes things like how many people were rushing the quarterback, and whether he was knocked down; the name of the pass defender in coverage on the play, and why a pass was incomplete; whether there was a scramble on the play, and who caused it, or if there was a sack, which offensive linemen blew their block and let the pass rusher through. It's an amazing resource, and now that it's entering its fifth year, the Game Charting Project is opening up countless new ways for us to analyze the game.

BNG: You have the Steelers 'projected average opponent' as 0.4% (20th) -- what does this mean?

BB: That means that the average projected DVOA amongst the Steelers' opponents in the upcoming season is 0.4%, the 20th-most difficult schedule amongst the league's 32 teams.

BNG: And the Pythagorean wins? I vaguely remember Pythagorous from 10th grade pre-calc, but I don't recall the theorum being applied to NFL football.

BB: The Pythagorean Theorem's usage in sports goes back nearly 30 years, and involves some now-pretty famous names.

In his series of Baseball Abstracts, Bill James made a series of discoveries that are still affecting the game of baseball to this day. One was that runs scored and runs against were a better predictor of a team's record in the following season than their win-loss record. James developed a formula similar to the Pythagorean Theorem to find a better predictor of wins, which he called a team's "Pythagorean Expectation".

The formula to determine a team's Pythagorean Expectation in baseball was Games * (Runs Scored2 / (Runs Scored2 + Runs Allowed2)).

About 20 or so years later, a researcher for STATS, Inc. started looking into the Pythagorean Expectation to see if it might be applicable to other sports, one of which was football. He found that the premise was applicable, with the results made more accurate by switching the exponent from 2 to 2.33. That researcher was Daryl Morey, who is now the General Manager of the Houston Rockets.

By replacing runs scored and runs allowed with points for and against, we can determine how many wins a team "should" have had in a given season. This figure is a superior indicator of wins in the following season than the team's win total. If a team has several more wins than its Pythagorean expectation, it's likely to fall back to earth in the following season.

There are two teams that stand out as obvious candidates to fall back in 2009 because of their Pythagorean expectation. The Dolphins won 11 games, but only outscored the opposition by a total of 28 points; that's the statistical signature of a team that wins, on average, 8.8 games. The Cardinals, meanwhile, went 9-7 despite scoring exactly one point more than they allowed. As you might expect, we'd figure them to be 8-8.

While a team's Pythagorean expectation is useful, team DVOA is actually a better predictor of wins in the subsequent season than either the team's win total in the previous year or their Pythagorean expectation. That's why we're so proud of it, of course.

07-30-2009, 11:45 PM
I'll gladly lay money on the "over."

07-31-2009, 06:11 AM
I guess it's interesting, but is all that math really necessary? I'd probably just listen to some illiterate bookie in Vegas.

So if the Steelers win 11 they're only off by 1.4 which they'll probably pat their backs for. Most of us can look at a team and get within 2 games what they'll end up by just using common sense.

If they're predicting the Lions to win 10 games, and they do... I'll be impressed.

07-31-2009, 10:22 AM
I'd take the over, as well. I doubt we'll have a 6.4 loss season. I think it'll be more of a 2.7 loss season, 3.8 at the most.


07-31-2009, 10:58 AM
This is the same kind of thinking behind such statements as "The Steelers offense is more effective with Melwelde Moore as the featured back".

It's a lot of fancy and complicated math, but is probably just about as accurate as tossing darts at a board...

07-31-2009, 11:18 AM
I'll gladly lay money on the "over."

Me too. I'm not sure if the simulation accounts for the fact that we might actually have a real running back this year.

Dino 6 Rings
07-31-2009, 11:22 AM
This is the same freaking website that I pointed out when used by some other person to debate Moore over Parker that has Ben as the 23rd Best QB in the League and ranked Nate Washington Higher than Santonio Holmes as Wide Receivers.

This Site is Fundementally Flawed.

07-31-2009, 11:57 AM
Before we jump on their prediction of only 9.6 wins, it's also worth noting that Football Outsiders has a higher probability of the Steelers finishing with 11 or more wins (37%) than winning nine or ten games (33%). Sounds to me like they're predicting at least 11-5, but that analysis may have been run before Baltimore found themselves with no receivers. Run the regression analysis again now and the most likely outcome may be up to 12-4.

Not quite the 14-2 mark I'm thinking of at this point, but their prediction isn't all that far off once you read the whole article and not just the headline.

07-31-2009, 12:01 PM
This is the same freaking website that I pointed out when used by some other person to debate Moore over Parker that has Ben as the 23rd Best QB in the League and ranked Nate Washington Higher than Santonio Holmes as Wide Receivers.

This Site is Fundementally Flawed.

I think their team stats for DVOA and DYAR are very good, but I'm much less impressed with any of their other stats as well as the individual stats.

07-31-2009, 12:39 PM
This team will go 12-4 to 14-2. Too much talent for them to lose more then that. If they do, then crap happens.

El-Gonzo Jackson
07-31-2009, 12:49 PM
This is the same kind of thinking behind such statements as "The Steelers offense is more effective with Melwelde Moore as the featured back".

It's a lot of fancy and complicated math, but is probably just about as accurate as tossing darts at a board...

Yeah, its the same thinking that had Football outsiders.com list the following guys as their top "Under the radar prospects for 2008"

Here are our top 10 prospects, lesser-known role players and brand-new starters who could make a difference in 2008 and/or beyond. (For this article, I've adjusted our list somewhat based on the preseason.)

1. Jerious Norwood, RB, Falcons
2. Jason Hatcher, DE, Dallas
3. Brandon McDonald, CB, Browns
4. Jerome Harrison, RB, Browns
5. Aaron Rouse, SS, Packers
6. Marques Harris, LB, Chargers
7. Ahmad Bradshaw, RB, Giants
8. Josh Beekman, OG, Bears
9. Tank Tyler, DT, Chiefs
10. Brent Celek, TE, Eagles

The Lakelander
07-31-2009, 02:57 PM
Just curious? How many wins did those nutbags predict we'd have last year? ... just sayin' :tt02:

07-31-2009, 04:03 PM
Im confused with the math but i dont care that some guy is ranking us...

07-31-2009, 05:07 PM
Im confused with the math but i dont care that some guy is ranking us...

For reals. I am sure the research is solid, but I just don't see the Steelers only winning 10 games this season with the schedule we were dealt.

07-31-2009, 07:36 PM
"The riddle of steel"

Ahhhhh...Thulsa Doom

08-01-2009, 04:08 PM
I think 9.6 sounds about right for us this year.... THROUGH WEEK 11!

08-01-2009, 05:02 PM
they still think the NFL season is 14 games.