Ok... to recap our defense needs to pick it up. Got it. But I can tell you put A LOT of time into those stats and it was pretty intresting. (Not being sarcastic)
Top Defensive Players of 2009
In December, I put up a post on the top Heisman candidates and my thoughts on them. With the emergence of Ndamukong Suh and locally with Brandon Graham, I wondered the best way to evaluate defensive players strictly from the stat sheet. Defense is made is more for the UFR, not for stat comparisons. The problem is, with over 800 games played in the FBS every year, it would take an army to break down the film for all the players in all the games. My stats based approach has the advantage of being able to quickly look at every game played last year.
There is no way to evaluate from the play by play who is responsible for a bad play on defense, but you can get a decent idea of who is responsible for a good one. Sure someone else could have opened up the hole, taken on extra blockers or forced a cutback, but over the course of a season, if you you made a lot key tackles, chances are you did a lot of work on those plays.
I took all of the plays from the season and immediately cut out all the plays from the second half where one team led by more than 2 TDs, no garbage time stat padding (same goes for any games against Baby Seal U opponents, always excluded from all my work). I then reduced the list of plays to ones that put the offense in a worse position than were they started the play. This doesn’t just mean TFL plays. A 2nd and 8 is worse than a 1st and 10 for an offense, so a tackle on a 2 yard gain on first down counts. Any third down stop counts, and often these are the biggest plays a defender can make. Turnovers are obviously the holy grail, stop the offense, create field position for your offense. The players are measured by two metrics, number of plays and magnitude of plays. A defensive tackle might make a lot of plays but most of them for relatively small values. A cornerback probably doesn’t get the chance to make many plays on a down by down basis, but each interception is huge, and has a very high value.
I then compared each players production versus what the average player at his position accomplishes to get a sort of VOAP, Value Over Average Player. Unfortunately, I couldn’t get good enough roster data to split all positions, so everyone is either DL, LB or DB, not perfect but better than nothing. I assumed that the average team would split the majority of the playing time between 6 DL, 4 LB and 5 DB. From the best I can tell, there isn’t much variation between the safeties and the corners, but DE’s get a bit of a benefit being compared to DT’s and DT’s get a slight hit compared to DE’s.
Top National Players
All player data is available here. It is easier to work with if you download into Excel since G Docs doesn’t like pages with a lot of rows. There are some slight changes to the numbers from my Heisman post as I reloaded the 2009 data and tweaked the expected value formulas.
I don’t have all the historical data but I would be shocked if a defensive tackle has ever had a better season than Ndamukong Suh did in 2009. He made 52 plays more and was nearly 5 TDs better than the average d-lineman and that includes defensive ends. No player in the country had a VOAP within a touchdown of him.
The closest, none other than Michigan’s Brandon Graham. Graham produced 27 extra plays and over 27 points of value more than the average defensive lineman. This made him the top value adding defensive end in the country and second to Suh overall.
In fact, the Big 10 had four of the top 6 defensive lineman in the country. O’Brien Schofield, Ryan Kerrigan and Adrian Clayborn all managed at least 23 points per page above replacement.
Luke Kuechly of Boston College led all linebackers with 57 plays and 25 points above average. Kuechly produced a ridiculous 87 negative plays on the season, 10 more than any other player at any position and 17 more than anyone else from a BCS a conference. Navorro Bowman, Nate Triplett and Brian Smith all cracked the bottom half of the top 10 linebackers at 17-18 points above average each.
Defensive backs were highly unproductive in the Big 10 relative to other conferences. Tyler Sash of Iowa came in at +12, 15th nationally and Donovan Warren was second best in the conference, but his +9 barely cracked the Top 40 nationally. Walter McFadden of Auburn was the top producing defensive back nationally, providing a +22 for the season.
Reviewing All-American Teams
I was curious to see how the national All-American team selections would compare with this metric. For positions like defensive end, linebacker and safety I would hope quite well because these positions are very output oriented and most of the value is in making plays. For defensive tackles and corner backs, I wasn’t as confident. A good defensive tackle will often add value by making plays for others. A good cover corner will often see the action go away from him and might not get many opportunities.
There were five players that were selected as defensive ends on the All-American teams, four of them stack up very well in my ratings, one does not. Brandon Graham was rated 1st and was +27. Derrick Morgan of Georgia Tech was 7th and +18. Von Miller of Texas A&M and Jason Pierre-Paul were 12th and 13th at about +15.5 for the season.
The outlier was TCU’s Jerry Hughes who came in 60th at +6 and was actually selected to the most All-American teams as any d-lineman. The large is probably due to the fact that 5 of his 11 sacks (10th nationally) came during garbage time.
Top non-selections were O’Brien Schofield, Wisconsin (2nd, +27), Jeremy Beal, Oklahoma (3rd, +25), Ryan Kerrigan, Purdue (4th, +25) and Adrian Clayborn, Iowa (5th, +23)
Again five players were picked to All-American teams as defensive tackles. Three had elite level production and 2 did not. Not to say that they were undeserving as discussed earlier, defensive tackles value can be difficult to attribute.
Suh was an obvious selection, and both Brian Price of UCLA (3rd, +20) and Gerald McCoy of Oklahoma (7th, +9) were well-deserved. The two potentially questionable selections, Mount Cody had the reputation and the highlight but did not have the direct productio, 79th nationally and 1.4 points below average. The other low production selection was Penn State’s Jared Odrick who just made the top 50 and was only 1.7 points above average.
Other top non-selections were Nate Collins, Virignia (2nd, +21), Lamarr Houston (4th, +16), Jared Crick, Nebraska (5th, +15) and Corey Peters, Kentucky (6th, +14).
There wasn’t much consensus among the All-American teams on the linebacker position. A total of 10 different selections were made. It appears the selections are more weighted on quantity of plays instead of quality of plays. This makes sense because most linebackers don’t make a lot of big stat sheet plays like interceptions or sacks and so the good old tackle stat is the most used.
When looking at the top values for linebackers, Luke Kuechly at #1 is the only player from the top 14 to receive any All-American honors. When you look at the play quantity, Kuechly and 3 others are in the top 8. The three are Rennie Curran from Georgia (+13), Pat Angerer, Iowa (+8) and Greg Jones from Michigan State (+12). Consensus pick Rolando McClain of Alabama is in the top 50 in both quantity and quality, and played for the top defense in the country.
There was one player who seemed to make the team purely on reputation and team success alone, because his production was dramatically less than the rest of the group. In defense of the selections, I won’t even name this individual for he might be the scariest man alive that played for the 2008 Florida National Championship game. He only produced 2 more plays and was just below average (-.1) in point production versus average player, and still received recognition from three different groups.
Of the five cornerbacks named to an All-American team in 2009, four landed in the Top 11 of my rankings. Joe Haden, Florida (3rd, +18), Javier Arenas (4th, +15), Alterruan Verner, UCLA (8th, +13) and Patrick Peterson from LSU (11th, +11) all produced very highly. The only exception was Perrish Cox from Oklahoma State who still managed to make the top 40 with a +6 for the season.
Walter McFadden from Auburn and Brandon Brinkley from Houston were the top two rated cornerback and both produced over +20 for the season.
It took seven selections to cover all of the picks for safeties. Five of the seven fit nicely at the top, including the top 3. DeAndre McDaniel, Clemson (1st, +20), Earl Thomas, Texas (2nd, +17), Rahim Moore, UCLA (3rd, +17), Tyler Sash, Iowa (7th, +12) and Eric Berry, Tennessee (11th, +10).
The two outliers were Kurt Coleman from Ohio State (44th, +1) and another apparent reputation selection, Taylor Mays, USC (77th, –3).
|Player||Position||Group||Plays||Value||Adj Plays||Adj Value|
|Ryan Van Bergen||DE||DL||23||16.1||5||6.5|
Michigan’s top two producers, Brandon Graham and Donovan Warren were covered above. After those two, only two players managed to be above +3 on the season. Ryan Van Bergen was +6.5 on the season on the defensive line and should have the potential for a big season this year. Stevie Brown was next at +6 (if you count him as a safety, +4 as a LB).
Jonas Mouton, Mike Martin, Obi Ezeh, Jordan Kovacs and Craig Roh all sit around the average mark for their positions. The most glaring point for me is that Michigan’s top linebacker, Mouton, barely makes the top 150 linebackers nationally in production. If Michigan’s defense is going to turn things around there is going to have be some new playmakers step up and there has to be more production from the linebackers.
Thanks for taking the time.
i wonder though if DB's are the position that gets dinged the most in a stats based analysis... the thought process of the ol' Deion Sanders / Charles Woodson theory, where if an offense is worried abotu a dominant DB, they just will not run plays in that player's direction, thereby artifically skewing his stats lower and artificially skewing the stats for his other side counterpart DB's higher (in quantity). Hence Eric Berry not being higher since teams would just not challenge him, fearing his play making ability. You can't do that type of offensive planning as much against DL's or LB's, but you can definately shade your passing game to single sides.
In anycase, I especially liked the analysis of Jerry Hughes. Again, I wonder though what the stats would look like if we bumped up the margin to 21 points instead of 14 points.
i think you could probably do a pretty good job sussing out who gets passed on the most if you looked at tackles from all pass plays.
is for you to do the work on the 1997 season.
This is a great test sample year because the Michigan Defense of 1997 produced a Heiseman Trophy winner on defense (yes some offense/punt returns in there for added TV value/exposure but my perception is that key turnover's are huge for Woodson)
But I also predict that many positions will be dominated by Michigan players, at least across the Big Ten.
From 1995 to 1997 I've never seen a more complete defense. And while I assume the volume of stats to prove this would be well beyond the hobby level of effort, I predict that teams that have above average DB's have above average DL and LB.
Afterall, if a team can't stop the run and short stuff, then the opposing offense does not need to risk the long ball. And this would show up statistically from enough data. At least I'm guessing.
Sorry for being an Oliver Twist, this is awesome stuff as it is,
but "please sir, can I have some more"
that a 1997 season analysis is the one that would prove if the formula is correct.
Really good post, but how is Boubacar Cissoko not even last.
Mathlete, I applaud your effort, but I don't know what the end results mean. Is there any way you could illustrate the various units of data used? How about Kenny Demens, for example. Since he was on the field for so few plays -- good, bad or irrelevant -- could you itemize the complete body of his work so we can see what constitutes his value rating (and what doesn't)?
The adjusted value measurement really only has value for full-time players. It compares their positive production versus the production of other full-time players at their position. Comparing Kenny Demens production on limited plays, isn't really valuable. What Kenny did do, was make 2 plays that put the opposing offense in a worse position. For his example, he was credited with a tackle on 1st and goal from the 1 against Wisconsin, that play had a value of .14, Wisconsin's expected points went from 6.34 on 1st and goal from the 1 to 6.20 on 2nd and goal from the 1. His second play was on a punt return. He combined with Kevin Leach to limit the returner to a 2 yard return after a 47 yard Mesko punt. The shorter return was worth .17 in field position for Michigan that was split between Leach and Demens.
I am interested in these value formulas, but have a few more questions. Where are you coming up with the information on expected points as it pertains to down and distance. I see the charts on expected points based on a 1st and 10 scenario, but how are you determining expected points for other down and distances (say for instance your 2nd and goal from the 1 that you use here)? Are there also charts for these? Or is there just a formula that you are using?
analogous to Brian's +1's from UFR? it would be worthwhile to add in negative plays, too, but that poses a much more difficult number crunch, right?
anyway, this may be part of what's happening with the LB numbers: you're always going to be able to measure the DL the best because there's nothing interfering with measuring their positive plays. they're the first line of defense. but the second line of defense is dependent on the first in making positive plays. so you should be able to adjust the LB numbers based off the DL numbers. i bet if you go back and run LB numbers with good DLmen and those with bad DLmen, you'll come up with an adjustment for opportunities to make those plays.
this is where the Mount Cody issue could be resolved to some degree too. Cody is a two-gap stuffer that forces plays away from him. there just aren't that many of those in the college game, so this kind of analysis just isn't suited for picking that kind of talent out...but it's also not a huge (pun intended!) issue. so there simply may not have been that many opportunities for McClain to make a ton of plays up the middle because there weren't a ton of plays run to the middle.
for defensive back, i would try to decide who doesn't get thrown at based on who makes more tackles on pass plays. you should be able at least to come up with an approximate number of times thrown at that that you can use as the denominator for INTs and PBUs.
obviously, this is a great job. and the work you did is kind of staggering. reminds me of building a fielding metric for baseball. there's usable stats out there, but better observations would be great too. that'd take an awful lot of coordination, though, and it's not like the NCAA has any interest in helping. it did indeed happen in baseball though, so there's hope.
Mike Williams on there. Was he just overlooked or am I missing something?