Patriots Defensive DVOA

[jimnance]

DVOA=trees.

Pts allowed=forest

 

[PatsFanInVa]

DVOA=trees.

Pts allowed=forest

Yardage allowed = er, nothing actually! LOL

Points allowed and points scored in the next 60-minute game = stat that matters.

Trouble is, that stat is always in the future

 

[jimnance]

Yardage allowed = er, nothing actually! LOL

Points allowed and points scored in the next 60-minute game = stat that matters.

Trouble is, that stat is always in the future

Problem is, people are trying to quantify this game with isolated stats, which they are doing in a vacuum.
That works to some degree with baseball and hoops, team sports that have a high degree of one- on- one matchups.
That is obviously not the case with the NFL. The sheer # of players on O and D, the multiplicity of formations, schemes, etc. not to mention that all these moving parts are not independent but rely on other moving parts working in sync. Especially a unit like the ol. So in NFL football, its about the gestalt. The #1 gestalt us W's followed by pts scored and pts allowed.
The best metric for D I've seen is Bruschis defensive index, which is weighted for most value given to pts allowed (minus pts scored by sts and opp def)with 3rd down, red zone, turnovers and rushing ydge. Passing yds allowed is basically meaningless for a winning team-since good teams often have large leads and they give up a lot of meaningless garbage-time passing yds. But for BAD teams, passing yds are NOT meaningless. So you can't apply a lot of stats to every team in the same way.

 

[denverpatsfan]

Giants defense is light years ahead of everyone else according to this ranking. So one would assume they can win ugly and make it to the Super Bowl.

Just how good is this Pats defense?

 

[stinkypete]

Any formula that has the current patriot defense rated 16th is out of touch with reality.

The DVOA measures performance across an entire season. The #16 ranking reflects 15 games played, the strength of schedule at the time of each game, etc. So, the defensive meltdown in Miami registers, in the DVOA formula, as a defensive collapse against a crappy team.

Adjusted DVOA adjusts strength of schedule to reflect the year end records rather than real-time records. Under adjusted DVOA, the Miami game hurts less because they are now considered a quality opponent. The adjusted DVOA also acknowledged that a team changes over the course of a year by weighting recent games more heavily, thus the jump from #16 to #6.

In any case, its an artificial and convoluted statistic used to generate web hits.

 

[Ring 6]

Problem is, people are trying to quantify this game with isolated stats, which they are doing in a vacuum.
That works to some degree with baseball and hoops, team sports that have a high degree of one- on- one matchups.
That is obviously not the case with the NFL. The sheer # of players on O and D, the multiplicity of formations, schemes, etc. not to mention that all these moving parts are not independent but rely on other moving parts working in sync. Especially a unit like the ol. So in NFL football, its about the gestalt. The #1 gestalt us W's followed by pts scored and pts allowed.
The best metric for D I've seen is Bruschis defensive index, which is weighted for most value given to pts allowed (minus pts scored by sts and opp def)with 3rd down, red zone, turnovers and rushing ydge. Passing yds allowed is basically meaningless for a winning team-since good teams often have large leads and they give up a lot of meaningless garbage-time passing yds. But for BAD teams, passing yds are NOT meaningless. So you can't apply a lot of stats to every team in the same way.

Here is the problem with ANY set of statistics or indexes or whatever.
First, there is nothing accounting for matchups, schemes and strengths and weaknesses.
The best team against the Giants may not be the best team against the Seahawks. The best offense against Pittsburghs defense may not be the best offense against Baltimores, or Oaklands.
Different offenses have different schemes and different players to execute those schemes. How a defense does vs the Patriots may have very little to do with how it does vs the Packers. Smaller quicker OLs will be best vs some defenses and worse vs others. etc, etc, etc.

Even after you get beyond that, you still end up with the fact that the game is about who scores the most points and there are 1000s of variables that go into who scores the most points that no statistical model could possibly account for, much less begin to assume how to properly weight.

If you want to determine who would win in computer simulations, then you can use stats, but on a real football field, DVOA doesn't exist when the ball is snapped.

 

[Ring 6]

The DVOA measures performance across an entire season. The #16 ranking reflects 15 games played, the strength of schedule at the time of each game, etc. So, the defensive meltdown in Miami registers, in the DVOA formula, as a defensive collapse against a crappy team.

Miami is a playoff team.
One half of bad defense while playing prevent with a rookie QB is your answer to why you are calling the #1 defense in the NFL in terms of ACCOMPLISHMENT average? Child please.

Adjusted DVOA adjusts strength of schedule to reflect the year end records rather than real-time records. Under adjusted DVOA, the Miami game hurts less because they are now considered a quality opponent. The adjusted DVOA also acknowledged that a team changes over the course of a year by weighting recent games more heavily, thus the jump from #16 to #6.

In any case, its an artificial and convoluted statistic used to generate web hits.

Its crap.

 

[PatsFanInVa]

Giants defense is light years ahead of everyone else according to this ranking. So one would assume they can win ugly and make it to the Super Bowl.

Just how good is this Pats defense?

Again - we'll find out in the last game they play. If it's the last game of the year - you know, the big game - and we win it... we might have a reallllly good Pats defense this year. In retrospect.

Those 2 SB defenses against the Gintz gave up 17 and 21 points respectively... and people here still say they "came up short." (and I'm going by the scores... I don't even remember if some of those points came on turnover/spec. teams plays).

This is a really good defense, to date.

If we make the SB, whoever we play - especially Dallas, an offensive juggernaut - if this defense goes in and pitches a shut-out... we'll look back and say, this was one of the great Pats defenses... if it goes on to smother teams for another year in the same fashion, people will start talking about one of the greatest ever, blah blah blah.

Or we can give up 30 points in the first round of the playoffs... again, that's the thing about stats.

 

[PatsFanInVa]

Problem is, people are trying to quantify this game with isolated stats, which they are doing in a vacuum.
That works to some degree with baseball and hoops, team sports that have a high degree of one- on- one matchups.
That is obviously not the case with the NFL. The sheer # of players on O and D, the multiplicity of formations, schemes, etc. not to mention that all these moving parts are not independent but rely on other moving parts working in sync. Especially a unit like the ol. So in NFL football, its about the gestalt. The #1 gestalt us W's followed by pts scored and pts allowed.
The best metric for D I've seen is Bruschis defensive index, which is weighted for most value given to pts allowed (minus pts scored by sts and opp def)with 3rd down, red zone, turnovers and rushing ydge. Passing yds allowed is basically meaningless for a winning team-since good teams often have large leads and they give up a lot of meaningless garbage-time passing yds. But for BAD teams, passing yds are NOT meaningless. So you can't apply a lot of stats to every team in the same way.

Your last line will probably be lost so I'll repeat it here

So you can't apply a lot of stats to every team in the same way.

(This was a comment on how various accumulations of stats would accrue differently to different teams, with the example of passing yards for teams always playing from behind.) Such teams would also have great defenses in the last 3 quarters if they are down by 35 points at the end of the first quarter

Style of play is just going to make a stat comparison difficult... the great defenses of all time each have their own character... some have HUGE numbers of takeaways... some just never let you score... etc. Of course the stats will overlap (the "smother" team will also be pretty good at turnovers).

In the BB case, he plays lots of soft zone when ahead, and everybody here hates it. Limit the big play, make it 10x tougher to get the score than the yards between the 20s.

But another thing to factor in is that stats have predictive value to the extent that play from one game will resemble play in the next game. The more innovative a team is, the more "ways a team has to beat you," the less predictive value a stat will have.

The stats should have great predictive value for a team like Seattle, which says "we'll just do what we do and challenge you to beat us."

I won't claim that BB is always the most innovative coach, but one hallmark is that the Pats will be ready with different game-plans for different opponents. They train to be ready to make adjustments based on situations. This limits the predictive value of "what worked last week" or "what they did in the first half," etc.

Okay, just a bunch of blather. I agree with you, is the upshot.

 

[zydecochris]

Here is the problem with ANY set of statistics or indexes or whatever.
First, there is nothing accounting for matchups, schemes and strengths and weaknesses.
The best team against the Giants may not be the best team against the Seahawks. The best offense against Pittsburghs defense may not be the best offense against Baltimores, or Oaklands.
Different offenses have different schemes and different players to execute those schemes. How a defense does vs the Patriots may have very little to do with how it does vs the Packers. Smaller quicker OLs will be best vs some defenses and worse vs others. etc, etc, etc.

Even after you get beyond that, you still end up with the fact that the game is about who scores the most points and there are 1000s of variables that go into who scores the most points that no statistical model could possibly account for, much less begin to assume how to properly weight.

If you want to determine who would win in computer simulations, then you can use stats, but on a real football field, DVOA doesn't exist when the ball is snapped.

Interesting post. I agree with you, but IMHO, there is another reason that a statistical analysis of football is more difficult and less effective than other sports: Sample Size. It is much easier to perform a meaningful statistical analysis of a large number of events than a few.

The sport that has been statistically analyzed for the longest time is baseball, where metrics such as batting average have been used for over a century. More recently, sophisticated empirical analysis of baseball statistics (sabermetrics) were used to build teams as described in "Moneyball". Why did this work so well with baseball? Well, there are 162 games per regular season and something like 146 pitches per game per team. Thus, there are over 46,000 plays per season, a very large sample size of plays, so lucky or unlucky "bounces of the ball" are mostly averaged out.

More recently, advanced statistical analysis has been used with basketball. Basketball has 82 games per regular season and something like 90 shots per game per team, so that is almost 15,000 plays per season, once again a fairly large sample size. Statistical analysis of basketball players and games has recently been used effectively also.

What about NFL football? There are only 16 games per regular season and about 65 snaps per team per game, which is only about 2000 plays per season. That is a much, much smaller sample size, and it means that "the way the ball bounces" is not averaged out over a season the way it does with baseball or basketball because of the smaller sample size. I'm sure that BB and his group of Rocket Scientists (Patricia) and brainiacs have their own statistics that they keep and specialized team analysis of those statistics, it isn't BB's way to not take every edge he can gain. But, as you point out, statistics are just not as reliable with football, I suspect that BB mostly relies on "the eye test" as he watches film.

 

[venecol]

If DVOA comes at the end of the entire season of what use is it? To compare with teams from previous years? Yet those teams wouldn't necessarily be the same due to personnel changes and injuries.

Seems like a complete waste of time.

 

[Ring 6]

Interesting post. I agree with you, but IMHO, there is another reason that a statistical analysis of football is more difficult and less effective than other sports: Sample Size. It is much easier to perform a meaningful statistical analysis of a large number of events than a few.

The sport that has been statistically analyzed for the longest time is baseball, where metrics such as batting average have been used for over a century. More recently, sophisticated empirical analysis of baseball statistics (sabermetrics) were used to build teams as described in "Moneyball". Why did this work so well with baseball? Well, there are 162 games per regular season and something like 146 pitches per game per team. Thus, there are over 46,000 plays per season, a very large sample size of plays, so lucky or unlucky "bounces of the ball" are mostly averaged out.

More recently, advanced statistical analysis has been used with basketball. Basketball has 82 games per regular season and something like 90 shots per game per team, so that is almost 15,000 plays per season, once again a fairly large sample size. Statistical analysis of basketball players and games has recently been used effectively also.

What about NFL football? There are only 16 games per regular season and about 65 snaps per team per game, which is only about 2000 plays per season. That is a much, much smaller sample size, and it means that "the way the ball bounces" is not averaged out over a season the way it does with baseball or basketball because of the smaller sample size. I'm sure that BB and his group of Rocket Scientists (Patricia) and brainiacs have their own statistics that they keep and specialized team analysis of those statistics, it isn't BB's way to not take every edge he can gain. But, as you point out, statistics are just not as reliable with football, I suspect that BB mostly relies on "the eye test" as he watches film.

I don't think BB gives a crap whether poor technique that let a receiver release inside, or a defender cross the face of an OL, or a player commit a penalty because he was out of position, means DVOA is affected. Football is not about accumulating statistics, or going back and evaluating a game based on what happened at what time and comparing it to other situations to come up with a cumulative total grade of plays. Football is about one on one battles, about getting the ball in the end zone, or keeping it out of the end zone, separating the ball carrier from the ball, exploiting the other teams weaknesses, calling the right play at the right time, and dozens of other things that don't show up in a geek analysis of stats.

 

[QuantumMechanic]

The DVOA measures performance across an entire season. The #16 ranking reflects 15 games played, the strength of schedule at the time of each game, etc.

Incorrect. DVOA takes into account the entire season (or season to date before the season is over). The piece of NE's DVOA that came from the first MIA game is based on all 15 MIA games, not just the two games MIA had played to that point.

 

[Ring 6]

Basically they give a score, based upon a scale they created, to each play, ten add up all those scores and say that these scores should determine who is a better football team.
They include things such as:



or this

Ultimately its an attempt to take a game that isn't about statistics and make it about statistics, so the flaws are inherent.

Quantum Mechanic is now disliking the definitions used by his beloved statistical model. LOL.

 

[QuantumMechanic]

Also, DVOA is an objective stat in that FO created a data model and then regressed it against actual win/loss records and chose weighting factors to give the best correlation with W/L. It's not some crap thing like ESPN's QBR.

 

[QuantumMechanic]

Quantum Mechanic is now disliking the definitions used by his beloved statistical model. LOL.

No, I'm disliking your asinine remarks about stats in general. You do realize that most or all NFL teams have analytics departments and FO (and other stats organizations your massive GM-ing experience leads you to dismiss out of hand) occasionally have consulting contracts with NFL teams.

 

[Ring 6]

No, I'm disliking your asinine remarks about stats in general. You do realize that most or all NFL teams have analytics departments and FO (and other stats organizations your massive GM-ing experience leads you to dismiss out of hand) occasionally have consulting contracts with NFL teams.

So you really think that NFL teams use crap like DVOA to tell them if their team is any good?
You think the Bill Belichick wants to know what a formula like DVOA tells him about his team?

I get that you think the game is played on a computer. I disagree.

 

[kkmack617]

DVOA is a much better stat than most and usually passes the eye test.

Defensive SRS is another one I like, Pats are currently #5.

Another one is defensive passer rating. Pats are #7, although I wish I could find one that adjusts for the opposing QB.

 

[Ring 6]

DVOA is a much better stat than most and usually passes the eye test.

Defensive SRS is another one I like, Pats are currently #5.

Another one is defensive passer rating. Pats are #7, although I wish I could find one that adjusts for the opposing QB.

Passer rating is a terrible stat. Its parameters are random and make no sense.
Anything that takes results and comes up with a way to mash up the good things for positive and the bad things for negative will generally get it somewhere in the vicinity of right.
But adding together a bunch of plays out of context with the idea that you can statistically make them judge football is ridiculous.

 

[Ring 6]

Also, DVOA is an objective stat in that FO created a data model and then regressed it against actual win/loss records and chose weighting factors to give the best correlation with W/L. It's not some crap thing like ESPN's QBR.

It is a crap thing. It places value on the steps to get to a result over the result. If you are the defense that allowed the fewest points per game, by over 2 a game, you are not the 16th most effective defense. You are only that if someone places more value on things that do not deserve them.
Next we will hear how Pro Football Focus is like the bible.