Post by Purple Pain on Dec 1, 2019 21:33:14 GMT -6
The Pro Bowl cases for the Vikings’ offense: What the advanced stats say by Arif Hasan
Diggs and Cook
Rest at link:
theathletic.com/1408995/2019/11/25/the-pro-bowl-cases-for-the-vikings-offense-what-the-advanced-stats-say/
Pro Bowl voting has started, and that’s just as good an excuse as any to look at the top players on the Minnesota Vikings and see where they fare compared to the rest of the league. This gives us an opportunity to take a look at the various forms of advanced statistics that have cropped up over the years and really dive into what people mean when they say “success rate,” “DVOA” or “adjusted net yards per attempt.”
In the last 10 years, the world of football analytics has exploded — and it can be difficult to keep up with the changes in how advanced statistics and football interact. The introduction of tracking data has magnified those challenges, which paradoxically helps statistics better reflect the reality of football while also making that data harder to understand. Sometimes, it can seem like attempts to decode football can make the whole process more confusing. We’ll take a look at many of those metrics and see what they say about current Vikings, and whether or not they point to a Pro Bowl-level season. Today, we’ll look at the offense.
Kirk Cousins
While it used to be the case that quarterbacks were largely judged by wins or total yards, we’ve seen significant growth in statistics that have provided a more accurate view of a quarterback’s performance. While single-season leaders in total yards tend to throw the ball a lot and throw the ball well, the players who place near the top in pass attempts often aren’t helping their teams as much as you’d think, generally throwing the ball a lot because their team is behind — a product perhaps of the poor quality of play coming from the quarterback position to begin with.
In the past, we’ve seen plenty of quarterbacks earn Pro Bowl selections in part because of high attempt and yardage totals, regardless of how efficient they were. Measuring by total yards, Cousins is seventh in the NFC, behind an interesting list of MVP candidates and anything but — Dak Prescott, Jameis Winston, Russell Wilson, Matt Ryan, Aaron Rodgers and Jared Goff.
One rate stat that attempts to make the picture clearer is passer rating, but it overvalues completion rate and over-penalizes interceptions while de-emphasizing touchdowns and yards and not being a particularly good indicator if win totals.
A widely adopted alternative is yards per attempt, which does a better job predicting future outcomes than passer rating does, but it ignores fairly significant events in a football game — sacks, turnovers and touchdowns.
In 1988, Bob Carroll, Pete Palmer and John Thorn took it a step further by developing an “adjusted” yards per attempt (AYA) that assigned 10 bonus yards for touchdowns and a 45-yard penalty for interceptions. After some additional work, Chase Stuart at the Pro-Football-Reference blog determined that the value of a touchdown is closer to 20 yards, which is the value that’s used now in the PFR leaderboards.
Using AYA, Cousins sits atop the NFC, with Wilson, Matthew Stafford and Prescott behind him.
But Jason Lisk at PFR found that quarterbacks control sack rates too, and therefore should be held responsible for them in the same way that they are yards, interceptions and touchdowns — after all, they don’t bear complete responsibility for those measures, either. More detailed analysis from Eric Eager at Pro Football Focus tells us that quarterbacks also have significant control over pressure rates.
In that case, evaluating quarterbacks in a way that includes sacks would, in theory, increase explanatory and predictive power. As Stuart found out at his website, Football Perspective, that turns out to be the case — “net” yards per attempt, which includes sack yardage lost as part of the measure and divides by total dropbacks instead of pass attempts, works better than yards per attempt and adjusted yards per attempt. And adjusted net yards per attempt works even better for explaining past performance.
By adjusted net yards per attempt, Cousins stays on top of the NFC, this time with Stafford right behind him, followed by Prescott and Wilson.
This, unfortunately, doesn’t include rushing yards, nor does it penalize quarterbacks for fumbles. Not only that, seven yards on third-and-8 are worth a lot less than those same yards on third-and-1. To include the impact of running the ball as well as the other type of turnover, Bryan Frye at GridFe developed a measure called total adjusted yards per play, or TAY/P.
This measure includes rushing and fumbles, adds a nine-yard bonus for first downs and excludes quarterback kneels and spikes. Here, fumbles are worth a 25-yard penalty because there’s a 50 percent chance that fumbles are recovered by the opposing team — but they still occur closer to the line of scrimmage, making them more damaging than interceptions, which is why the penalty is more than half that of an interception.
By this measure, Cousins is fourth in the NFC, behind Prescott, Wilson and Stafford.
But with additional computing capability comes stronger measures of success. Knowing that an interception deep downfield on third-and-15 is a lot less impactful than a short interception on first-and-goal — and that some teams will score just as much at the goal line running the ball as they will passing it — we can actually calculate the specific value each quarterback has added on each individual play based on down, distance and field position.
Aaron Schatz at Football Outsiders developed a metric based on this understanding, incorporating a concept called “success rate” that Carroll, Palmer and Thorn pioneered in their book. In this case, any play on first down that generates at least 45 percent or more of the required yardage needed to convert is a success. On second down, that percentage changes to 60 percent, while third and fourth down predictably require at least 100 percent of the yardage to convert to be considered successful.
There are marginal values for non-successful plays and bonuses for explosive plays, as well as those in the red zone. After normalizing those values against the rest of the league, the result was “value over average,” with roughly half of players (or teams) above average and the other half below. Then, Schatz made team adjustments (as well as adjustments for weather and so on), resulting in “defense-adjusted value over average,” or DVOA. Using this method, Cousins ranks fourth in the NFC behind the usual suspects.
One can apply defensive adjustments to ANYA as well, which Justis Mosqueda of Optimum Scouting has done, placing Cousins fourth in the same way DVOA has. (Note: TAYP, DVOA and defense-adjusted ANYA are only updated through Week 11.)
The play-by-play measurement method was actually something that Virgil Carter, former Bengals quarterback, was able to generate back in 1971 for a measure he called expected points added. Every down, distance and field position has an “expected points” value, which is contingent upon the average number of points a team will score from that situation. First-and-10 at your own 10-yard line, for instance, is worth a lot less than the same down and distance at your opponent’s 10-yard line.
The statistic essentially calculates the difference in each expected point total from play to play and assigns that difference to the offense — and in this case, also to the quarterback on plays where he throws or runs.
The problem is that it had been difficult to consistently and easily produce expected points for each quarterback on a week-by-week basis because of the challenges involved with parsing out play-by-play data. Brian Burke, who later joined ESPN to work with their Stats & Info department, hosted expected points tables at his website, Advanced Football Analytics but then ported them over to ESPN’s Total QBR page.
Pro-Football-Reference also has expected points, which one can find for individual players using their play finder. With that tool, Cousins ranks second in the NFC in expected points, just behind Prescott and ahead of Stafford and Wilson.
ESPN’s tool is a bit different, however, and parses out drops, air yards — that is the distance the ball travels in the air on completions, in contrast to yards after the catch — garbage time, pressured plays as well as a few other corrections they detail in their explainer.
With those additional corrections to EPA and a normalizing effect to distribute player scores to a range between 0 and 100, with 50 as average, ESPN’s QBR is one of the most context-friendly statistics out there. In QBR, Cousins ranks fourth in the conference, behind the three quarterbacks we’ve been mentioning.
Tracking data has added even more, with the ability to evaluate every throw and determine how probable a completion is based on a variety of factors, like how many yards in the air the ball travels, how close the nearest defender to the receiver is, the distance of the receiver from the sideline, pressure on the quarterback and more.
By adding up all the completions and incompletions a quarterback generates and subtracting it from how many an average quarterback would complete based on those throwing conditions, we get what Next Gen Stats calls “completions plus-minus.” A more publicly accessible version called “completion percentage over expected” (CPOE) doesn’t use the tracking data and instead uses throw distance to determine the “expected” completion rate. That’s why The Athletic’s Ben Baldwin combined EPA and CPOE to create DAKOTA, a measure that predicts future EPA even better than current EPA does.
In CPOE, Cousins ranks first in the NFC, ahead of the rest of the gang-of-four NFC QBs, Wilson, Prescott and Stafford. In DAKOTA, he remains atop the NFC. If we count Drew Brees, who has played six games this season, Cousins ranks second in CPOE and DAKOTA.
But if we want to look at the full context — determine who’s responsible for catches or incompletions, figure out where fumbles come from and whether or not a quarterback threw a good ball that led to yards after-the-catch or benefitted from a screen pass and so on — we probably need film grading.
The folks at Pro Football Focus have done that, grading every player on every play. Contrary to what many think they do, they grade based on the footage, not statistics. By determining whether a player did their job effectively, they assign a grade between -2.0 and +2.0 with zero given the vast majority of time for plays where an average player would have performed the same job.
That way, quarterbacks are not punished for drops or rewarded for off-target throws that receivers catch through remarkable acrobatics. They will record statistics, but those statistics are a product of their grading process and don’t inform the grades themselves.
They then turn those raw numerical grades into a normalized 0-100 system, like ESPN QBR. According to PFF, Cousins ranks fourth in the NFC — ahead of Prescott and Stafford but falling behind Rodgers and Brees as well as Wilson.
Add all that up and average out the various quarterbacks’ ranks in these stats, and here’s where Cousins stands:
In the last 10 years, the world of football analytics has exploded — and it can be difficult to keep up with the changes in how advanced statistics and football interact. The introduction of tracking data has magnified those challenges, which paradoxically helps statistics better reflect the reality of football while also making that data harder to understand. Sometimes, it can seem like attempts to decode football can make the whole process more confusing. We’ll take a look at many of those metrics and see what they say about current Vikings, and whether or not they point to a Pro Bowl-level season. Today, we’ll look at the offense.
Kirk Cousins
While it used to be the case that quarterbacks were largely judged by wins or total yards, we’ve seen significant growth in statistics that have provided a more accurate view of a quarterback’s performance. While single-season leaders in total yards tend to throw the ball a lot and throw the ball well, the players who place near the top in pass attempts often aren’t helping their teams as much as you’d think, generally throwing the ball a lot because their team is behind — a product perhaps of the poor quality of play coming from the quarterback position to begin with.
In the past, we’ve seen plenty of quarterbacks earn Pro Bowl selections in part because of high attempt and yardage totals, regardless of how efficient they were. Measuring by total yards, Cousins is seventh in the NFC, behind an interesting list of MVP candidates and anything but — Dak Prescott, Jameis Winston, Russell Wilson, Matt Ryan, Aaron Rodgers and Jared Goff.
One rate stat that attempts to make the picture clearer is passer rating, but it overvalues completion rate and over-penalizes interceptions while de-emphasizing touchdowns and yards and not being a particularly good indicator if win totals.
A widely adopted alternative is yards per attempt, which does a better job predicting future outcomes than passer rating does, but it ignores fairly significant events in a football game — sacks, turnovers and touchdowns.
In 1988, Bob Carroll, Pete Palmer and John Thorn took it a step further by developing an “adjusted” yards per attempt (AYA) that assigned 10 bonus yards for touchdowns and a 45-yard penalty for interceptions. After some additional work, Chase Stuart at the Pro-Football-Reference blog determined that the value of a touchdown is closer to 20 yards, which is the value that’s used now in the PFR leaderboards.
Using AYA, Cousins sits atop the NFC, with Wilson, Matthew Stafford and Prescott behind him.
But Jason Lisk at PFR found that quarterbacks control sack rates too, and therefore should be held responsible for them in the same way that they are yards, interceptions and touchdowns — after all, they don’t bear complete responsibility for those measures, either. More detailed analysis from Eric Eager at Pro Football Focus tells us that quarterbacks also have significant control over pressure rates.
In that case, evaluating quarterbacks in a way that includes sacks would, in theory, increase explanatory and predictive power. As Stuart found out at his website, Football Perspective, that turns out to be the case — “net” yards per attempt, which includes sack yardage lost as part of the measure and divides by total dropbacks instead of pass attempts, works better than yards per attempt and adjusted yards per attempt. And adjusted net yards per attempt works even better for explaining past performance.
By adjusted net yards per attempt, Cousins stays on top of the NFC, this time with Stafford right behind him, followed by Prescott and Wilson.
This, unfortunately, doesn’t include rushing yards, nor does it penalize quarterbacks for fumbles. Not only that, seven yards on third-and-8 are worth a lot less than those same yards on third-and-1. To include the impact of running the ball as well as the other type of turnover, Bryan Frye at GridFe developed a measure called total adjusted yards per play, or TAY/P.
This measure includes rushing and fumbles, adds a nine-yard bonus for first downs and excludes quarterback kneels and spikes. Here, fumbles are worth a 25-yard penalty because there’s a 50 percent chance that fumbles are recovered by the opposing team — but they still occur closer to the line of scrimmage, making them more damaging than interceptions, which is why the penalty is more than half that of an interception.
By this measure, Cousins is fourth in the NFC, behind Prescott, Wilson and Stafford.
But with additional computing capability comes stronger measures of success. Knowing that an interception deep downfield on third-and-15 is a lot less impactful than a short interception on first-and-goal — and that some teams will score just as much at the goal line running the ball as they will passing it — we can actually calculate the specific value each quarterback has added on each individual play based on down, distance and field position.
Aaron Schatz at Football Outsiders developed a metric based on this understanding, incorporating a concept called “success rate” that Carroll, Palmer and Thorn pioneered in their book. In this case, any play on first down that generates at least 45 percent or more of the required yardage needed to convert is a success. On second down, that percentage changes to 60 percent, while third and fourth down predictably require at least 100 percent of the yardage to convert to be considered successful.
There are marginal values for non-successful plays and bonuses for explosive plays, as well as those in the red zone. After normalizing those values against the rest of the league, the result was “value over average,” with roughly half of players (or teams) above average and the other half below. Then, Schatz made team adjustments (as well as adjustments for weather and so on), resulting in “defense-adjusted value over average,” or DVOA. Using this method, Cousins ranks fourth in the NFC behind the usual suspects.
One can apply defensive adjustments to ANYA as well, which Justis Mosqueda of Optimum Scouting has done, placing Cousins fourth in the same way DVOA has. (Note: TAYP, DVOA and defense-adjusted ANYA are only updated through Week 11.)
The play-by-play measurement method was actually something that Virgil Carter, former Bengals quarterback, was able to generate back in 1971 for a measure he called expected points added. Every down, distance and field position has an “expected points” value, which is contingent upon the average number of points a team will score from that situation. First-and-10 at your own 10-yard line, for instance, is worth a lot less than the same down and distance at your opponent’s 10-yard line.
The statistic essentially calculates the difference in each expected point total from play to play and assigns that difference to the offense — and in this case, also to the quarterback on plays where he throws or runs.
The problem is that it had been difficult to consistently and easily produce expected points for each quarterback on a week-by-week basis because of the challenges involved with parsing out play-by-play data. Brian Burke, who later joined ESPN to work with their Stats & Info department, hosted expected points tables at his website, Advanced Football Analytics but then ported them over to ESPN’s Total QBR page.
Pro-Football-Reference also has expected points, which one can find for individual players using their play finder. With that tool, Cousins ranks second in the NFC in expected points, just behind Prescott and ahead of Stafford and Wilson.
ESPN’s tool is a bit different, however, and parses out drops, air yards — that is the distance the ball travels in the air on completions, in contrast to yards after the catch — garbage time, pressured plays as well as a few other corrections they detail in their explainer.
With those additional corrections to EPA and a normalizing effect to distribute player scores to a range between 0 and 100, with 50 as average, ESPN’s QBR is one of the most context-friendly statistics out there. In QBR, Cousins ranks fourth in the conference, behind the three quarterbacks we’ve been mentioning.
Tracking data has added even more, with the ability to evaluate every throw and determine how probable a completion is based on a variety of factors, like how many yards in the air the ball travels, how close the nearest defender to the receiver is, the distance of the receiver from the sideline, pressure on the quarterback and more.
By adding up all the completions and incompletions a quarterback generates and subtracting it from how many an average quarterback would complete based on those throwing conditions, we get what Next Gen Stats calls “completions plus-minus.” A more publicly accessible version called “completion percentage over expected” (CPOE) doesn’t use the tracking data and instead uses throw distance to determine the “expected” completion rate. That’s why The Athletic’s Ben Baldwin combined EPA and CPOE to create DAKOTA, a measure that predicts future EPA even better than current EPA does.
In CPOE, Cousins ranks first in the NFC, ahead of the rest of the gang-of-four NFC QBs, Wilson, Prescott and Stafford. In DAKOTA, he remains atop the NFC. If we count Drew Brees, who has played six games this season, Cousins ranks second in CPOE and DAKOTA.
But if we want to look at the full context — determine who’s responsible for catches or incompletions, figure out where fumbles come from and whether or not a quarterback threw a good ball that led to yards after-the-catch or benefitted from a screen pass and so on — we probably need film grading.
The folks at Pro Football Focus have done that, grading every player on every play. Contrary to what many think they do, they grade based on the footage, not statistics. By determining whether a player did their job effectively, they assign a grade between -2.0 and +2.0 with zero given the vast majority of time for plays where an average player would have performed the same job.
That way, quarterbacks are not punished for drops or rewarded for off-target throws that receivers catch through remarkable acrobatics. They will record statistics, but those statistics are a product of their grading process and don’t inform the grades themselves.
They then turn those raw numerical grades into a normalized 0-100 system, like ESPN QBR. According to PFF, Cousins ranks fourth in the NFC — ahead of Prescott and Stafford but falling behind Rodgers and Brees as well as Wilson.
Add all that up and average out the various quarterbacks’ ranks in these stats, and here’s where Cousins stands:
Diggs and Cook
Rest at link:
theathletic.com/1408995/2019/11/25/the-pro-bowl-cases-for-the-vikings-offense-what-the-advanced-stats-say/
... and neither is this Chris guy.