- Joined
- Feb 9, 2014
- Messages
- 20,922
- Name
- Peter
The Rams Are Young And Talented: Will It Matter?
By Frank "Dubs" Dobozy @Bozy1313 on Aug 20 2014
Jeff Curry-USA TODAY Sports
The future looks bright for the St. Louis Rams. They're expected to rise above the .500 mark - and challenge for a playoff berth - this coming season. Now into year three of a complete rebuild under Jeff Fisher and Les Snead, the time has arrived for the Rams' young talent to turn potential into results on the field.
Since the arrival of Jeff Fisher and Les Snead in early 2012, theRams have been quite successful in stockpiling talented young players. Football Outsiders/ESPN - in its annual ranking of NFL teams with the most under-25 talent (link) - named the St. Louis Rams as the NFL team with the best collection of young players.
The future appears bright for the Rams. They're expected to rise above the .500 mark - and challenge for a playoff berth - this coming season. Now into year three of a complete rebuild under Fisher and Snead, the time has arrived for the Rams' young talent to turn potential into results on the field.
In a recent article for FiveThirtyEight [owned by ESPN], Neil Paine attempted to determine:
"...whether there’s a relationship in the NFL between having a lot of young talent and winning down the road."
After investigation and analysis, Paine came to a conclusion that may surprise you:
"...there’s practically no relationship between how highly regarded a team’s talent is and how well it does in the next three seasons."
Paine relies heavily on statistical analysis in his article. I like to think of myself as a numbers guy; however, most of the statistical measures Paine uses - Spearman rank correlation coefficient, schedule-adjusted Pythagorean wins, Defense-Adjusted Yards Above Replacement - leave me scratching my head in wonderment.
In an article for Turf Show Times last week, I shared my thoughts on the ever-increasing use of advanced NFL statistics:
"I look at all of these new-fangled NFL stats - particularly the ones involving a fair amount of subjectivity - with a degree of skepticism. Statistics can tell you everything...and nothing. This is especially true for statistics related to NFL football. Individual statistics are invariably interconnected - with many moving parts - making it difficult to assess talent and performance levels."
Wes Cecil is a valued, respected contributor and member of the Turf Show Times community. He's written many quality fanposts, where he's displayed a great interest in NFL statistical analysis. Wes is a professor of literature and philosophy, and teaches mathematics logic, critical thinking, and the foundations of scientific reasoning. When not working, reading, or writing, Wes spends much of his time in the garden with his chickens.
I asked Wes to contribute his thoughts on the limitations of statistical analysis with respect to the NFL, and to offer his commentary on Neil Paine's article. It makes for a compelling read:
By Frank "Dubs" Dobozy @Bozy1313 on Aug 20 2014
Jeff Curry-USA TODAY Sports
The future looks bright for the St. Louis Rams. They're expected to rise above the .500 mark - and challenge for a playoff berth - this coming season. Now into year three of a complete rebuild under Jeff Fisher and Les Snead, the time has arrived for the Rams' young talent to turn potential into results on the field.
Since the arrival of Jeff Fisher and Les Snead in early 2012, theRams have been quite successful in stockpiling talented young players. Football Outsiders/ESPN - in its annual ranking of NFL teams with the most under-25 talent (link) - named the St. Louis Rams as the NFL team with the best collection of young players.
The future appears bright for the Rams. They're expected to rise above the .500 mark - and challenge for a playoff berth - this coming season. Now into year three of a complete rebuild under Fisher and Snead, the time has arrived for the Rams' young talent to turn potential into results on the field.
In a recent article for FiveThirtyEight [owned by ESPN], Neil Paine attempted to determine:
"...whether there’s a relationship in the NFL between having a lot of young talent and winning down the road."
After investigation and analysis, Paine came to a conclusion that may surprise you:
"...there’s practically no relationship between how highly regarded a team’s talent is and how well it does in the next three seasons."
Paine relies heavily on statistical analysis in his article. I like to think of myself as a numbers guy; however, most of the statistical measures Paine uses - Spearman rank correlation coefficient, schedule-adjusted Pythagorean wins, Defense-Adjusted Yards Above Replacement - leave me scratching my head in wonderment.
In an article for Turf Show Times last week, I shared my thoughts on the ever-increasing use of advanced NFL statistics:
"I look at all of these new-fangled NFL stats - particularly the ones involving a fair amount of subjectivity - with a degree of skepticism. Statistics can tell you everything...and nothing. This is especially true for statistics related to NFL football. Individual statistics are invariably interconnected - with many moving parts - making it difficult to assess talent and performance levels."
Wes Cecil is a valued, respected contributor and member of the Turf Show Times community. He's written many quality fanposts, where he's displayed a great interest in NFL statistical analysis. Wes is a professor of literature and philosophy, and teaches mathematics logic, critical thinking, and the foundations of scientific reasoning. When not working, reading, or writing, Wes spends much of his time in the garden with his chickens.
I asked Wes to contribute his thoughts on the limitations of statistical analysis with respect to the NFL, and to offer his commentary on Neil Paine's article. It makes for a compelling read:
"It turns out you really don't need to understand statistics to see the central flaw in Paine's argument: Junk In = Junk Out. Paine's measure of success over time of the most talented young teams is not based on some fundamental data, but on ESPN's and Football Outsider's ranking of the quality of the under-25 talent on each team:
"Instead of limiting our analysis to players who have yet to emerge in the NFL, these rankings consider all players who will be 25 or younger as of September 1, 2010 -- regardless of where they were drafted or how many games they've started. After compiling a list of eligible players for each team, we compared the groups on a variety of factors. We weighed issues like upside versus established production, quantity versus quality, and current staff versus historical ability to develop rookies when it comes to evaluating the talent available to each NFL franchise."
This is a highly subjective analysis. What Paine is actually arguing is not that there is no correlation between young talent and future success, but that there is no correlation between ESPN's ability to predict the performance of young players and the future success of associated teams - a somewhat less surprising conclusion. This means that if, on the off chance, ESPN has graded one of your team's young stars much higher than they should be, their failure to contribute in the future will make your team look bad. Similarly, if they think your team's young talent is terrible, then any future success looks anomalous.
We should not be too harsh on ESPN, Football Outsiders or Paine. As Paine mentions in his article, "in general, the NFL is much more difficult to predict than MLB." I think it is important to understand why all this apparently heavy-duty statistical analysis falls apart when examining the NFL. Consider, for example, that the 2010 Chargers led the league in total offense and total defense. They also failed to make the playoffs.
This extraordinary achievement tells us, above all, that the total offense and total defense statistics are not very good measures of team success. For statistical analysis to be helpful, you need three elements that are missing from the NFL - a low number of variables, reliable quantities, and a large sample.
The NFL has an astounding number of variables. Not only are there 22 players on the field on each play, those players vary from play to play. There are huge differences in offensive and defensive schemes that are being matched, games are played in temperature controlled stadiums on artificial turf and in the snow on grass. Essentially, every game involves so many variables in personnel, scheme and conditions it is difficult to compare one game to another.
Many of the variables in football are also difficult to quantify. A player like Calvin Johnson forces a defense to adjust to him. Even if he does not catch a ball for long stretches of a game, he is influencing the game decisively. In the video for Robert Quinn being chosen as one of the NFL's top 100 players, Coach Pagano of the Colts is seen on the sideline saying, "we said we weren't going to let Quinn ruin our game plan." The Colts had adjusted what they were doing to account for Quinn, and he still ruined their game. When Johnson does not get a catch or Quinn doesn't get a sack, it looks like a poor statistical game. However, if you constantly draw double and triple-teams, you have a major impact. But how do you put a number on that?
And if you cannot put a reliable number on something, then it cannot be accounted for in statistics. Junk In = Junk Out. There are simply too many variables, and many of those variables are hard to quantify, for reliable use of statistics. Pro Football Focus tries to account for all of this by grading every player on every play of every game. However, as they readily admit, such grades are subjective and - because of an incomplete knowledge of what a player was supposed to do - often inaccurate.
Another problem for statistical analysis of the NFL is the sample size. Because there are so many variables, it is difficult to get a sufficient number of similar samples to create reliable base-line data. In baseball there are 2,430 regular season games a year. In football, 256, or roughly 1/10 the number of games. Further, far more players participate in each game.
Even with relief pitchers and designated runners, likely only 30 to 40 players participate in a given baseball game. In an average football game, you will likely see more than 80 players take the field. So 1/10 the number of games and at least twice as many players means a nearly worthless sample size for many of the variables we would like to understand.
For instance, in evaluating the talent of a young slot corner, how many snaps is he likely to see in a year? If your team plays 30 defensive plays a game, and you are in nickel 30% of the time, our young slot corner might see 160 snaps in a season. If a team runs half the time against the nickel, and targets the slot receiver 20% of the time when they throw, our player will see 16 passes in his direction in an entire season. That is an astonishingly low number to derive any reliable data.
Our corner could fall down on a play and allow a 90 yard touchdown, get picked on a play and allow a 45 yard completion, and then stop every other pass for no gain. That would equal an 8.4 yard completion average. Is that good or bad? It is an ugly number, but a great year for a slot corner. Because of the number of variables, it is often impossible to have sufficient number of samples to get clear base-line data.
Further exacerbating the situation, the NFL changes the rules of the game every year. Why have pass catching tight-ends suddenly risen while running backs have seen their importance decline? Rule changes. Strong safeties used to roam the backfield praying for a QB to throw a seam pass to a TE or slot receiver. The strong safety would then destroy the receiver. Receivers would develop alligator arms and offensive coordinators who wanted to extend the lives of their receivers would avoid throwing to the middle of the field.
Now, devastating blows are called "hits on defenseless receivers" and result in 15-yard penalties and league fines. Suddenly, the middle of the field is open. Similarly, emphasis on hand checking and pass-interference has also changed the dynamics of the passing game. These rule changes mean that from year to year, and certainly over any five year period, the game has changed sufficiently, that it is difficult to compare statistics across time.
Of all the difficulties of analysis presented by the NFL, perhaps none is more daunting than player evaluation. Most NFL teams employ seven or eight full time scouts, plus player development directors and managers and scouting services. Further, when the season ends, the entire coaching staff becomes scouts and begins evaluating players, attending and running workouts, and watching film.
These guys watch a lot of film. So twenty or more men spend thousands of hours analyzing players in order to choose seven or so players in the draft, a handful of UDFA's, and sign maybe a half-dozen free-agents. Despite all of this work, players are drafted early who fail to perform and players are drafted late or come in as UDFA's and make a real contribution. Is it any surprise the team at ESPN and Football Outsiders are unable to more accurately predict the future success of young players, than the accumulated talent of the NFL's best minds?
So when you see statistical analysis of the NFL be very leery. When Chip Kelly was hired as Philadelphia's head coach, several articles discussed his use of a computer system called Zeus, that had used advanced statistical analysis to come up with the idea that you should almost always go for it on 4th and short.
This analysis suggested, by the way, that all the other head football coaches in the NFL did not understand the game. Kelly, the argument ran, was going to revolutionize football by going for it a LOT more on 4th down. As it turns out, in his first year as head coach, Chip Kelly called 14 fourth down plays and was successful on 7 of them. This was essentially middle of the pack in number called and success rate.
The NFL cannot be accurately quantified because of the large number of variables, subjective nature of assigning values to many variables, and the limited number of samples available for many of the variables that could be quantified. This does not mean that some hard numbers are not helpful in giving insight to the game or player production, rather that any such analysis must be mated to a careful viewing of each and every game relevant to the question at hand. As coaches say repeatedly at the end of games, "I'll let you know when we have finished watching the tape."
"Instead of limiting our analysis to players who have yet to emerge in the NFL, these rankings consider all players who will be 25 or younger as of September 1, 2010 -- regardless of where they were drafted or how many games they've started. After compiling a list of eligible players for each team, we compared the groups on a variety of factors. We weighed issues like upside versus established production, quantity versus quality, and current staff versus historical ability to develop rookies when it comes to evaluating the talent available to each NFL franchise."
This is a highly subjective analysis. What Paine is actually arguing is not that there is no correlation between young talent and future success, but that there is no correlation between ESPN's ability to predict the performance of young players and the future success of associated teams - a somewhat less surprising conclusion. This means that if, on the off chance, ESPN has graded one of your team's young stars much higher than they should be, their failure to contribute in the future will make your team look bad. Similarly, if they think your team's young talent is terrible, then any future success looks anomalous.
We should not be too harsh on ESPN, Football Outsiders or Paine. As Paine mentions in his article, "in general, the NFL is much more difficult to predict than MLB." I think it is important to understand why all this apparently heavy-duty statistical analysis falls apart when examining the NFL. Consider, for example, that the 2010 Chargers led the league in total offense and total defense. They also failed to make the playoffs.
This extraordinary achievement tells us, above all, that the total offense and total defense statistics are not very good measures of team success. For statistical analysis to be helpful, you need three elements that are missing from the NFL - a low number of variables, reliable quantities, and a large sample.
The NFL has an astounding number of variables. Not only are there 22 players on the field on each play, those players vary from play to play. There are huge differences in offensive and defensive schemes that are being matched, games are played in temperature controlled stadiums on artificial turf and in the snow on grass. Essentially, every game involves so many variables in personnel, scheme and conditions it is difficult to compare one game to another.
Many of the variables in football are also difficult to quantify. A player like Calvin Johnson forces a defense to adjust to him. Even if he does not catch a ball for long stretches of a game, he is influencing the game decisively. In the video for Robert Quinn being chosen as one of the NFL's top 100 players, Coach Pagano of the Colts is seen on the sideline saying, "we said we weren't going to let Quinn ruin our game plan." The Colts had adjusted what they were doing to account for Quinn, and he still ruined their game. When Johnson does not get a catch or Quinn doesn't get a sack, it looks like a poor statistical game. However, if you constantly draw double and triple-teams, you have a major impact. But how do you put a number on that?
And if you cannot put a reliable number on something, then it cannot be accounted for in statistics. Junk In = Junk Out. There are simply too many variables, and many of those variables are hard to quantify, for reliable use of statistics. Pro Football Focus tries to account for all of this by grading every player on every play of every game. However, as they readily admit, such grades are subjective and - because of an incomplete knowledge of what a player was supposed to do - often inaccurate.
Another problem for statistical analysis of the NFL is the sample size. Because there are so many variables, it is difficult to get a sufficient number of similar samples to create reliable base-line data. In baseball there are 2,430 regular season games a year. In football, 256, or roughly 1/10 the number of games. Further, far more players participate in each game.
Even with relief pitchers and designated runners, likely only 30 to 40 players participate in a given baseball game. In an average football game, you will likely see more than 80 players take the field. So 1/10 the number of games and at least twice as many players means a nearly worthless sample size for many of the variables we would like to understand.
For instance, in evaluating the talent of a young slot corner, how many snaps is he likely to see in a year? If your team plays 30 defensive plays a game, and you are in nickel 30% of the time, our young slot corner might see 160 snaps in a season. If a team runs half the time against the nickel, and targets the slot receiver 20% of the time when they throw, our player will see 16 passes in his direction in an entire season. That is an astonishingly low number to derive any reliable data.
Our corner could fall down on a play and allow a 90 yard touchdown, get picked on a play and allow a 45 yard completion, and then stop every other pass for no gain. That would equal an 8.4 yard completion average. Is that good or bad? It is an ugly number, but a great year for a slot corner. Because of the number of variables, it is often impossible to have sufficient number of samples to get clear base-line data.
Further exacerbating the situation, the NFL changes the rules of the game every year. Why have pass catching tight-ends suddenly risen while running backs have seen their importance decline? Rule changes. Strong safeties used to roam the backfield praying for a QB to throw a seam pass to a TE or slot receiver. The strong safety would then destroy the receiver. Receivers would develop alligator arms and offensive coordinators who wanted to extend the lives of their receivers would avoid throwing to the middle of the field.
Now, devastating blows are called "hits on defenseless receivers" and result in 15-yard penalties and league fines. Suddenly, the middle of the field is open. Similarly, emphasis on hand checking and pass-interference has also changed the dynamics of the passing game. These rule changes mean that from year to year, and certainly over any five year period, the game has changed sufficiently, that it is difficult to compare statistics across time.
Of all the difficulties of analysis presented by the NFL, perhaps none is more daunting than player evaluation. Most NFL teams employ seven or eight full time scouts, plus player development directors and managers and scouting services. Further, when the season ends, the entire coaching staff becomes scouts and begins evaluating players, attending and running workouts, and watching film.
These guys watch a lot of film. So twenty or more men spend thousands of hours analyzing players in order to choose seven or so players in the draft, a handful of UDFA's, and sign maybe a half-dozen free-agents. Despite all of this work, players are drafted early who fail to perform and players are drafted late or come in as UDFA's and make a real contribution. Is it any surprise the team at ESPN and Football Outsiders are unable to more accurately predict the future success of young players, than the accumulated talent of the NFL's best minds?
So when you see statistical analysis of the NFL be very leery. When Chip Kelly was hired as Philadelphia's head coach, several articles discussed his use of a computer system called Zeus, that had used advanced statistical analysis to come up with the idea that you should almost always go for it on 4th and short.
This analysis suggested, by the way, that all the other head football coaches in the NFL did not understand the game. Kelly, the argument ran, was going to revolutionize football by going for it a LOT more on 4th down. As it turns out, in his first year as head coach, Chip Kelly called 14 fourth down plays and was successful on 7 of them. This was essentially middle of the pack in number called and success rate.
The NFL cannot be accurately quantified because of the large number of variables, subjective nature of assigning values to many variables, and the limited number of samples available for many of the variables that could be quantified. This does not mean that some hard numbers are not helpful in giving insight to the game or player production, rather that any such analysis must be mated to a careful viewing of each and every game relevant to the question at hand. As coaches say repeatedly at the end of games, "I'll let you know when we have finished watching the tape."
