Wednesday, January 30, 2013

Data Analysis: Win Probability Added

For those that saw Moneyball or are familiar with Sabremetrics in baseball, you know the usefulness and application of some methods of advanced statistics. Baseball is a statisticians dream as you can whittle down every play, every moment in the game to a series of individual matchups. Pitcher vs hitter. Hitter vs fielder. Hitter vs park. The possibilities are endless and an entire industry around advanced statistical analysis has grown around baseball. Other sports, like football, are a little harder to analyze because of the team aspect of the game. The fact that there are 11 moving parts on each side of the ball all at the same time can mean that an offensive lineman correctly executed his block on a defender, but because of a blocking failure elsewhere on the line, the play was not a success. Football also has a much smaller sample size to draw from than baseball. Football has about one-tenth the games (16 opposed to 162) and about half as many plays per game (baseball games see close to 300 pitches per game while football games see about 130 plays per game).

One of the best sites for football statistics is Advanced NFL Stats, who developed win probability graphs by logging data on down, distance, field position, score and time remaining. They run graphs for each game that can show at any given point in a game the probability of a team winning. From this general model, they are able to isolate individual players and determine a Win Probability Added (WPA) value.
Win Probability Added (WPA) – The difference between a team’s Win Probability (WP) at the start of a play and the WP at the end of the play. WPA is the measure of a play’s impact on the outcome of a game. An individual player’s WPA is the sum of the WPA of the plays in which that player was directly involved. Being directly involved is defined as an offensive player who ran, threw, or kicked the ball, was targeted by a pass, or flagged for a penalty. Defensive players are credited for WPA when they tackle or sack the ball carrier, are credited with an assisted tackle or sack, cause a fumble, defend a pass, or are flagged for a penalty. 
This creates an opportunity to compare players across different teams or even within the same team to determine the approximate value they added to the team. As the definition says, on offense only skill position players qualify for their WPA metric. Season WPAs are determined by adding the WPA from each game played over the course of the season. The one problem with this is that players that miss portions of the season due to injury (such as LaMarr Woodley or Troy Polamalu) would almost automatically have lower WPAs than players who play the whole season. Therefore it is also useful to look at WPA per game.

On offense, Ben Roethlisberger led the way with a 2.25 WPA, which ranked him 14th amongst quarterbacks and 17th amongst all offensive players. When digging deeper into the stats, we see that Ben's numbers were built up over the first half of the year before the wheels fell off the wagon.
Interestingly, Ben's numbers started to take a turn for the worse in the Giants game, not after he was injured in the Chiefs game. This adds some credence to the notion that it was actually Antonio Brown's injury that was the turning point in the season. After Brown's injury against the Giants, not only did Ben's numbers take a dip, but the offense as a whole seemed to sputter and die. To put things in perspective, through 8 games (including the Giants game), Ben had a WPA of 3.11, which put him on pace for a 6.2 WPA for the season. This would have put Ben head and shoulders above every other quarterback in the league. Aaron Rodgers finished the season with the top WPA of an offensive player with 5.02 followed by Matt Ryan (4.87) and Tom Brady (4.69). Even if Ben had simply maintained his pace over the second half of the season and averaged no change over the last 8 games, he would have finished in 10th, ahead of RGIII, Russell Wilson, and Joe Flacco. As it was, the interceptions against Dallas and Cincinnati were season-killers for the Steelers, and the numbers bear that out.

For the receivers, Brown and Sanders actually posted very similar WPA per game numbers. Sanders actually had the highest WPA of the receivers because he was the only one of the group to appear in all 16 games. The most striking number amongst the receiver group is Mike Wallace's -0.78 number. This ranked him 76th amongst receivers in the league. Game-by-game breakdowns for receivers aren't available yet, but looking at Wallace's career numbers we can see some distinct changes in 2012.

Wallace saw declines in several key categories, including Success Rate, Yards Per Target, Catch Rate, and Deep Ball Percentage. For this measure, Deep Ball Percentage was defined as the percentage of targets over 15 yards downfield. While this has been steadily declining with each year of his career, the more telling statistic is Catch Rate, where Wallace caught only 53.8% of balls that were thrown his way this year, as opposed to 60.5% last season. Without this 7% decrease, Wallace would have had 72 receptions in 15 games. Even working with just his pure average for yards per catch, he would have had 105 more yards on the season. You can make any kind of excuses you want for Wallace's numbers being down, such as less deep balls, less targets (though his targets per game was actually the highest of his career) or less yards per target. But at the end of the day, the biggest number on the chart is the Catch Rate. The struggles of Mike Wallace's hands were well-documented this year and that is the biggest reason for his decline in production.

On the defensive side of the ball, the Steelers featured 5 of the top 60 defenders in the league and had 4 players rank in the top 10 at their position. Additionally, 7 players ranked in the top 30 in the league at their position, which is significant considering the "linebacker" category lumps together both inside and outside linebackers. According to the WPA stat (and backed up by anyone who watched the Steelers this year), Lawrence Timmons was the top player on the defense. Timmons ended the year with a WPA of 2.36, which was not only the best on the team, but also the best amongst all linebackers (both inside and outside). The most amazing stat is that Timmons' 2.36 WPA was second in the league behind only Houston's JJ Watt who posted an amazing 2.96.

Timmons had a great year, there's no denying that. How he was once again shafted from getting an invitation to the Pro Bowl is just plain crazy. But looking at the bigger picture, this wasn't even the best year of Timmons' career from a WPA standpoint. In 2010, Timmons posted a 2.65 WPA, which was actually higher than Defensive Player of the Year Troy Polamalu's 2.43 and James Harrison's 2.14 in 2008 when he won DPOY. Speaking of 2010, there are some stark differences that jump off the page when you compare the linebacking play from 2010 with 2012.

In 2012, only Timmons and Foote played all 16 games. The outside linebacker positions were a shuffle between Harrison, Woodley, Worlids and Carter. As a whole, the group only accounted for 25 sacks and 46 quarterback hits along with 40 tackles for loss. On the turnover side, the linebackers only forced 6 fumbles and had 4 interceptions.

The first thing you notice when looking at the 2010 data is that all four starting linebackers played in all 16 regular season games. This consistency showed in WPA where 3 of the 4 had WPA's over 2 and Farrior's 1.91 was a close 4th. Compare that to 2012 where the second most productive linebacker (Harrison) had a WPA of 1.70. That's a huge difference in impact plays from the linebacking corps. As far as other raw data, the 2010 group accounted for 33 sacks and 50 QB hits. The 2010 corps also produced turnovers, forcing 12 fumbles and coming down with 5 interceptions.

To compare the two groups, the 2010 squad had 8 more sacks. The difference here was the production of Harrison and Woodley. In 2010 they combined for 20 sacks while in 2012 they only combined for 12. This difference in 8 sacks is magnified when you also consider the 2010 group only had 4 more QB hits. This means the linebackers were getting pressure with approximately the same consistency but not getting home and landing sacks when they got close. The problems with getting to the quarterback cascaded down to turnovers where the Steelers forced half as many fumbles this year as they did in 2010. The dropoff once again can be directly pinpointed to Harrison and Woodley. In 2010, Harrison forced 6 fumbles, primarily through his tomahawk chop move coming around the back side of the quarterback. However, Harrison recorded 5 fewer sacks in 2012 than in 2010 and had 4 fewer forced fumbles. This is not to place the entire burden of the season on the linebackers, but there were significant deficiencies, particularly when compared with the last season in which the Steelers won the division and reached the Super Bowl.

In 2010, no one on the defensive line had a WPA over 1, but in 2012 Brett Keisel posted a 1.39, which was actually the 10th best WPA for a defensive end in the league. This is significant because Advanced NFL Stats groups 3-4 DEs and 4-3 DEs together. This means that Keisel scored higher than a significant number of 4-3 pass rushing DEs that tallied double the amount of sacks that Keisel did.

Despite having the #1 pass defense again this season, the Steelers secondary saw a significant dropoff from 2010 to 2012 in terms of WPA.

In 2010, Troy Polamalu's Defensive Player of the Year performance accounted for a 2.41 WPA in just 14 games. On the corners, Ike, B-Mac and William Gay all played 16 games. Gay led the way with the highest WPA, with McFadden (who split time with Gay throughout the season) coming in at 1.09.

That same 1.09 number that McFadden posted in 2010 as the second-highest CB was the highest WPA by a Steelers corner in 2012. Cortez Allen, whose numbers got a huge boost thanks to his play over the last two games of the season when he forced 3 fumbles and recorded 2 interceptions, was the highest rated Steelers corner. This does not mean he was the best player on the field, but the boost he got from those turnovers propelled him up the charts. Allen and Keenan Lewis's WPAs ranked them in the top 25 corners in the league. Lewis finished the season second in the league behind Seattle's Richard Sherman in passes defended. However, he was the only defensive back in the top 10 in the league in passes defended to not record an interception.

By comparison, Sherman and Jennings are mentioned in the Defensive Player of the Year discussion. Intercepting passes is part of becoming a complete corner. Lewis has proven he can get a jump on the ball and make plays to break up passes. As he develops, he's going to have to start coming down with interceptions if he wants to make good on his promise of being a Pro Bowler.

Looking back at WPA, the biggest improvement obviously came from Ryan Clark who had a 0.78 in 2010 and improved to a secondary-leading 1.42 in 2012. In fact, Clark's 2012 season was the 6th best amongst safeties in the league. The loss of Troy Polamalu for half of the season obviously hurt the secondary, and it showed in the turnover numbers. In 2010 the secondary had 14 interceptions while in 2012 they only had 6.

All told, when data from the Steelers last Super Bowl run is held up side-by-side with their 2012 season, the biggest story is injuries. In 2010 the top 4 linebackers, top 3 corners, and 1 safety played in all 16 games. Fast forward to 2012 and the only players on defense to play in all 16 games were Lawrence Timmons and Larry Foote. On offense, only Emmanuel Sanders played in all 16 games this season, and he battled through some injuries through the middle part of the season. In 2010, the Steelers got contributions in all 16 games from Rashard Mendenhall, Mike Wallace and Hines Ward. On the offensive line, the Steelers started 9 different linemen this year with only Ramon Foster and Max Starks starting all 16 games. Back in 2010, the Steelers had 8 different linemen start with Maurkice Pouncey and Flozell Adams starting all 16 games and Chris Kemoeatu starting 15.

Whatever personnel decisions the Steelers make moving forward, the most important thing for success is to keep players healthy and on the field. We saw across  the board how injuries wrecked havoc on WPA from Ben Roethlisberger's game-by-game numbers bottoming out after Antonio Brown's injury and Ben's subsequent injury to the outside linebackers who were mixing and matching pairings all season. If there is a "silver bullet" out there that can fix the Steelers it would be the ability to stay healthy and injury-free. The Steelers aren't deficient in talent and still have top-tier players on both offense and defense. The potential is there to be right back in the mix in 2013 to compete for another Lombardi Trophy. The talent is there, the team just has to stay healthy and make the necessary plays down the stretch to bring home wins.

1 comment:

Gd Girish said...

The official standardised survey is excellent when you want to examine and gauge theory and do Data cleaning. It will certainly frequently have actually predetermined inquiries with matching meanings or explanations to make sure that recruiters and participants will certainly not be confused. Official surveys will also have pre-set feedback style to facilitate fast completion of the survey during the interview process. An exploratory questionnaire design is the opposite of the official one.