If you haven't done so yet, you'd better read Approximate Value I. (I know you're thinking, "yeah, yeah, whatever." I do the same thing. But I mean it. Go read the old post.)
I left off last time with this bunch of questions:
- What metric do I use to determine offensive points at the team level?
- What fraction of points should go to the line?
- What is the pass/run split?
- On the passing side, what is the throw/catch split?
- We need to figure a way to give some of those offensive line points to fullbacks and tight ends, many of whose jobs include a lot of blocking.
Hopefully, I'll be able to answer all these today, and run through an example or two. Before I do, I'd like to make a few comments about the method and about my style of doing these sorts of things:
- I'll just state upfront that this is a case where I'm not necessarily opposed to tweaking the metric until it gives us results we're happy with, instead of picking a theoretical basis and forcing ourselves to stick with it. As I quoted Bill James in the last post: "These approximations are not intended to tell you anything at all about the player that you do not already know." They're not supposed to teach us new things; they're merely supposed to codify the things we already know, so it's OK to cook the books a little bit until they do tell us what we already know. The problem here is that none of us really knows how to compare Tarik Glenn's 2006 to Gary Clark's 1991. And to the extent that we do "know," we all "know" different things. The point is: while I do think we need some sort of theory to get us started in certain areas, I won't be too apologetic about making some arbitrary changes if a strict application of the theory leads us to "wrong" answers.
- The reason I write a blog instead of writing books --- well, one of them --- is because I'm not the kind of guy who thinks everything through completely before running the numbers and writing it up. I'm the kind of guy who comes up with a vague idea, gets excited about it, and tries to get to (and share) some preliminary results as soon as possible, occasionally making a few admittedly half-hearted choices along the way. There are some calculations below that I know are wrong. But I think they're close enough that they won't do any real damage to the conclusions. I'll go back and re-examine them later.
- A reminder that this method is for the purpose of establishing approximate value. From yesterday's post, this is James: "The approximations are intended only to distinguish as quickly and reliably as possible between large contributions, very large contributions, gigantic contributions, medium-sized contributions, small, smaller, and negligible contributions." I bolded "as possible" to remind us that there's only so much we're going to be able to do. We're not going to be able to give Hines Ward the credit he allegedly deserves for blocking. We're not going to be able to distinguish between a tight end who posted 20/110/1 because he stunk and another who posted 20/110/1 because he was basically a third tackle. If a team has a very good right guard and a very bad left guard, we're not going to be able to distinguish them unless the good one makes the pro bowl. Is this unfortunate? Yes. Does it make this whole exercise useless? That's for you to decide. But if you think it does, there's not much point in reading on. So if you're still here, remember: I will liberally and sometimes arbitrarily play the "we're keeping it simple" card. Daniel Graham and Steve Tasker will simply have to accept my apologies.
- At least for now, I'm only going to attempt to rate players from 1978--2006 (I'm still filling in a few stats from '07). First, lots of things changed with the rule changes that preceded the 1978 season. They have continued to change since then, of course, but I'm fairly comfortable assuming that 78--06 constitutes an era. Also, games started will be a key part of the metric, and GS data gets a little sketchy in my database right around 1978.
OK, here goes.
The metric for measuring team offense
At least for now, I'm going to go with offensive points per drive, which is a stat I should have thought to compute a long time ago, but didn't. Unless I'm missing something, every drive ends with either a rushing touchdown, a passing touchdown, a field goal attempt, a punt, a turnover, a failed fourth down conversion attempt, or the end of a half. On the team level, the new database has touchdowns, field goals, punts, and turnovers. The other two I don't have data for. A half-ending possession that doesn't result in one of the four main outcomes probably wasn't much of a possession anyway, so I don't mind not counting those. I wish I had turnover-on-downs data, but given the relative rarity of these kind of possessions, I'm not too uncomfortable excluding them. So we'll just call it offensive points per estimated drive:
OPPED = ( 7*TDs + 3*(field goals made) ) / ( TDs + FG attempts + punts + turnovers )
Now divide the team's OPPED by the league OPPED, multiply by 100, and that's how many points a team's offense has to distribute. So an average offensive team will necessarily have 100 points. The 1982 Chargers, 2000 Rams, and 1984 Dolphins all have about 170. The 1992 Seahawks have about 44 and the 2006 Raiders have 47.
Dividing up the points
Here's where it's going to get a little controversial.
In the comments to yesterday's post, Neil outlined an idea that is probably better than mine:
I tried to look at salaries to see how GMs value each position. For instance, QBs made 17% of the salary cap # devoted to offensive players from 2000-07, so I allocated 17% of “Offensive Wins” to the QB position; RBs were paid 12% so they got 12% of wins, etc
If I had that data, and if I'd thought of it, that's probably what I'd use. But I don't and I didn't. I did, however, realize that last April I ran a study that can provide a similar theoretical basis for allocating the points. I calculated the percentage of draft value chart "points" that NFL teams have historically used on each position. Here is the chart:
qb 7.4 rb 12.5 wr 11.7 te 3.9 ol 15.5 dl 19.3 lb 13.2 db 15.8 pk 0.5 pn 0.3
Assuming teams actually believe the pick value chart, and assuming they know what they're doing, then it might make sense to make it a goal to set up our system so that, in the long run, the number of total points awarded to players at each position matches the above distribution. We're going to start with that idea and work from there.
I wish I had a RB/FB breakdown, but I don't. I'm just going to make one up: 80/20. So we have this:
qb 7.4 rb 10.0 fb 2.5 wr 11.7 te 3.9 ol 15.5
Looking at just QB, RB, WR, TE, and OL, it adds to 51. Our task is essentially to divide that 51 into four buckets: (1) blocking, (2) running, (3) pass throwing, (4) pass catching. Let's make the following assumptions, which I do know are not exactly right:
- an OL's job is 100% blocking
- an average QB's job is 95% pass-throwing and 5% running
- an average RB's job is 70% running and 30% pass-catching
- a WR's job is 100% pass-catching (sorry Hines)
- an average TE's job is 70% pass-catching and 30% blocking (yes, I know this varies a lot from TE to TE, more on that later)
- an average fullback's job is 10% running, 20% catching, and 70% blocking.
So, for blocking we have 100% of 15.5, 30% of 3.9, and 70% of 2.5.
For running we have 5% of 7.4, 70% of 10.0, and 10% of 2.5.
For pass-throwing we have 95% of 7.4.
For pass-catching we have 100% of 11.7, 30% of 10.0, 20% of 2.5, and 70% of 3.9.
It all adds up to:
So the proportions are:
I really, really like the above method. The problem is that it just doesn't seem to work. Offensive linemen turn out to be undervalued by just about anyone's standards. In a preliminary version of this method using these percentages, Jon Ogden came in right between Brian Sipe and Ronnie Harmon in terms of total career approximate value. So let's tweak up the blocking percentage just a bit and keep the remaining relationships fixed. Here's one that I think works pretty well.
Blocking: 45.5% <--- that's 5/11
OK, so an average team will have 45.5 points to split among the blockers. The 1982 Chargers will have about 77. The 2006 Raiders will have about 22. Here's how we'll split it:
- Every lineman, fullback, and tight end gets 1 pre-point for each game played and an additional 5 pre-points for each game started.
- Tackles get their pre-points multiplied by 1.3, fullbacks by .7, and tight ends by .3.
- Pro bowl linemen (not tight ends or fullbacks) get their pre-points multiplied by 1.7.
- Every OL, FB, and TE gets points proportional to his percentage of the team's total pre-points.
Now we move on to the skill guys. The percentages above dictate that an average team should have 23.5% of its remaining points devoted to running, and 76.5% devoted to throwing and catching. [Here come a couple of calculations that I think are basically in the ballpark, but could use some improvement...] For each team, we take its ratio of rushing yards to total yards and divide it by the league average ratio. This gives a number like 1.15 for a run-heavy team or .93 for a team with a slight tendency toward the pass. Then we multiply that number by 23.5 to see what percentage of that team's non-blocking points will go to runners. Now we know how many points to give the runners, and how many points to give the passers/receivers.
Remember from last time, I declared that the passer/receiver split should stay constant from team to team. So passers get 11.8/(11.8+29.9) = 28.3% of the passing game points and receivers get the other 71.7%. Remember, there are a lot more people that have to split the receiving points.
The individual runners get points proportional to their share of the team's rushing yards. We're keeping it simple here. Likewise with the individual passers (passing yards) and receivers (receiving yards).
That's it. Now let's look at some results and see if it's believable. Here are the top two and bottom two offensive teams of the era, along with a couple of average-ish teams:
sdg 1982 172.8 -------------- Dan Fouts 21 Wes Chandler 20 Kellen Winslow 19 Doug Wilkerson 15 Russ Washington 14 Billy Shields 14 Chuck Muncie 13 Ed White 10 Charlie Joiner 10 Don Macek 10 James Brooks 7 Eric Sievers 5 John Cappelletti 2 Pete Holohan 2 Chuck Loewen 2 Andrew Gissinger 2 Bob Rush 2 Dwight Scales 2 Bobby Duckworth 1 Dennis McKnight 1 Scott Fitzkee 1 mia 1984 171.6 -------------- Dan Marino 22 Mark Clayton 15 Ed Newman 14 Dwight Stephenson 14 Jon Giesler 14 Mark Duper 14 Cleveland Green 11 Tony Nathan 11 Roy Foster 10 Dan Johnson 9 Woody Bennett 6 Nat Moore 6 Bruce Hardy 5 Joe Carter 5 Eric Laakso 3 Joe Rose 3 Jeff Toews 2 Jimmy Cefalo 2 Pete Johnson 2 Jim C. Jensen 2 Ronnie Lee 2 Steve S. Clark 1 Andra Franklin 1 nyj 2001 100.1 -------------- Curtis Martin 15 Vinny Testaverde 11 Laveranues Coles 9 Kevin Mawae 8 Jason Fabini 8 Ryan Young 8 Wayne Chrebet 7 Richie Anderson 7 Kerry Jenkins 6 Anthony Becht 5 Randy Thomas 4 LaMont Jordan 3 J.P. Machado 2 Kevin Swayne 2 James Dearth 1 David Loverne 1 Jerald Sowell 1 oti 2000 99.7 -------------- Eddie George 14 Steve McNair 12 Brad Hopkins 10 Frank Wycheck 9 Bruce Matthews 8 Derrick Mason 8 Fred Miller 7 Benji Olson 5 Chris Sanders 5 Kevin Long 5 Erron Kinney 3 Lorenzo Neal 2 Yancey Thigpen 2 Carl Pickens 2 Neil O'Donnell 2 Rodney Thomas 1 Kevin Dyson 1 Zach Piller 1 Mike Leach 1 Jason Mathews 1 rai 2006 47.0 -------------- Justin Fargas 4 Doug Gabriel 4 Ronald Curry 4 Langston Walker 4 Jake Grove 3 Aaron Brooks 3 Kevin Boothe 3 Randy Moss 3 Andrew Walter 3 Robert Gallery 2 Zack Crockett 2 LaMont Jordan 2 Paul McQuistan 2 Courtney Anderson 2 Chad Slaughter 2 Barry Sims 2 ReShard Lee 1 Alvis Whitted 1 John Madsen 1 Corey Hulsey 1 Randal Williams 1 sea 1992 43.2 -------------- John L. Williams 6 Chris Warren 5 Stan Gelbaugh 3 Andy Heck 3 Ray Roberts 3 Kelly Stouffer 2 Darrick Brilz 2 Tommy Kane 2 Bill Hitchcock 2 Joe Tofflemire 2 Paul Green 1 Ronnie Lee 1 Louis Clark 1 John Hunter 1 Robb Thomas 1 Ron Heller 1 James R. Jones 1 Brian Blades 1 Trey Junkin 1 David Daniels 1
Here's the total career approximate value list. Players whose careers started before 1978 (and hence whose complete careers are not counted) are asterisked:
Dan Marino 225 Jerry Rice 223 Brett Favre 212 John Elway 207 Bruce Matthews 176 Anthony Munoz 176 Warren Moon 172 Emmitt Smith 171 Steve Young 165 Joe Montana 161 Lomas Brown 157 Marshall Faulk 156 Vinny Testaverde 154 Peyton Manning 153 Mike Kenn 152 Drew Bledsoe 147 Boomer Esiason 146 Barry Sanders 141 Willie Roaf 140 Shannon Sharpe 139 Gary Zimmerman 139 Dave Krieg 138 Marcus Allen 136 *Jackie Slater 136 Richmond Webb 136 Will Shields 134 Marvin Harrison 134 Randall Cunningham 133 Curtis Martin 133 Thurman Thomas 132 Jim Kelly 132 *Dan Fouts 132 Steve McNair 132 Randall McDaniel 131 Troy Aikman 130 Tim Brown 129 James Lofton 128 Mark Brunell 126 Rich Gannon 125 Orlando Pace 125 Tarik Glenn 123 Bruce Armstrong 122 *Mike Webster 122 Stan Brock 122 Art Monk 121 Phil Simms 121 Cris Carter 121 Ricky Watters 120 Andre Reed 120 Henry Ellard 120 Jim Everett 119 Steve DeBerg 119 Will Wolford 117 Tony Gonzalez 117 *Walter Payton 116 Michael Irvin 116 Edgerrin James 116 Chris Hinton 114 John L. Williams 114 *Steve Largent 113 Rod Smith 112 Max Montoya 112 Terrell Owens 112 Joe Jacoby 111 Jonathan Ogden 111 Tiki Barber 110 Isaac Bruce 109 Steve Wisniewski 109 Tom Nalen 109 *Tony Dorsett 109 Jerome Bettis 109 Jimmy Smith 108 Ozzie Newsome 108 Jim Harbaugh 108 Walter Jones 108 Luis Sharpe 108 Eric Dickerson 107 Bob Whitfield 107 Trent Green 107 Tony E. Jones 106 Mike Munchak 106 Kerry Collins 105 Brad Hopkins 105 Willie Anderson 105 Jake Plummer 104 Chris Chandler 104 Warrick Dunn 104 Roger Craig 104 Todd Steussie 102 Earnest Byner 102 *Joe Theismann 101 Larry Centers 100
The best single seasons of the period form an interesting collection that will surely spark some discussion:
Orlando Pace 2000 23 Marshall Faulk 1999 23 LaDainian Tomlinson 2006 23 Dan Marino 1984 22 Priest Holmes 2002 22 Steve Young 1993 22 Steve Young 1992 21 Tarik Glenn 2004 21 Dan Fouts 1982 21 Terrell Davis 1998 21 Steve Young 1994 21 Emmitt Smith 1995 21 Marshall Faulk 2000 20 Peyton Manning 2004 20 Steve Wallace 1992 20 Peyton Manning 2006 20 Daunte Culpepper 2004 20 Wes Chandler 1982 20 Edgerrin James 1999 20 Kurt Warner 2001 20 Barry Sanders 1994 20 Rich Gannon 2000 20 Daunte Culpepper 2000 20 Kellen Winslow 1982 20
Finally, here are the approximate top performers of 2006:
LaDainian Tomlinson 2006 23 Peyton Manning 2006 20 Tarik Glenn 2006 19 Steven Jackson 2006 17 Tiki Barber 2006 17 Antonio Gates 2006 17 Philip Rivers 2006 17 Larry Johnson 2006 17 Marcus McNeill 2006 16 Marvin Harrison 2006 16 Flozell Adams 2006 16 Reggie Wayne 2006 15 Carson Palmer 2006 15 Matt Light 2006 15 Michael Vick 2006 15 Tom Brady 2006 15 Frank Gore 2006 15 Drew Brees 2006 15 Jon Kitna 2006 14 Brian Westbrook 2006 14 Willie Anderson 2006 14 Jeff Saturday 2006 14 Jammal Brown 2006 14 Joseph Addai 2006 14 Marc Bulger 2006 14 Chad Johnson 2006 13 Chris Samuels 2006 13 Chad Pennington 2006 13
You may agree with these lists in spots and strongly disagree in others. That's OK. But it's worth pausing for a minute to remind ourselves just how impossible the task we've undertaken is. All we're using is passing yards, rushing yards, receiving yards, games, games started, position played, and pro bowl status. And we're only using the pro bowls when we really have to (for linemen). Given that that's all we have to work with, I happen to think the method does a pretty good job of getting close to an assessment of approximate value that most people would generally agree with.
Kinds of players who are over-valued:
- On the career list, quarterbacks who played for a long time. That's unavoidable, given that QB is the most important position on the field, quarterbacks tend to have long careers, and this is a career total metric. But it does raise some questions. Namely, if a team has a terrible offense and/or a terribly inefficient passing game, why are we giving the quarterback any points at all? Stan Gelbaugh got three points for compiling dreadful stats on the worst offensive team of the last 30 years. Why aren't we giving him negative points? I'm not sure I have a good answer for that, except that I don't see how to do it in a way that would keep the method for QBs consistent with the rest of the positions and I'm not quite ready to have totally different sets of rules for different positions. And I definitely don't want to go the route of making this a metric where an average player is zero and a below average player has negative points.
- Post-James Wilder era running backs, compared to pre-James Wilder era backs. Wilder was the first true workhorse back. Before him, no coach ever attempted to let a single RB get 80--90% (or more) of the RB work on a team. I'm not completely comfortable with Shaun Alexander having three seasons as good as Earl Campbell's best one, simply because Tim Wilson was getting some touches.
- Bad offensive tackles. Of course, they won't generally be on good teams and/or won't stick around long anyway. So I don't see this as a major issue.
- Offensive linemen who make the pro bowl based on reputation long after their best years are behind them.
- Offensive linemen (especially tackles) who were the only pro bowl lineman on a very good offensive team. Orlando Pace and Tarik Glenn have probably been overcredited a little.
- Guys whose nominal position is fullback, but who don't do as much blocking as a real fullback. Yes, Mike Alstott, I'm talking about you. Probably Larry Centers too. They get as big a share of the blocking credit as Dan Kreider and Lorenzo Neal do, but they don't do near as much actual blocking.
I was hoping to avoid this, but I may have to re-work the running game / passing game percentages, creating a slightly different set for pre-1985ish versus post-1985ish.
Kinds of players who are under-valued:
- Good blocking tight ends and wide receivers.
- Linemen (especially guards and tackles) who were good, but not good enough to make the pro bowl.
Next, we move to the defensive side of the ball. That may happen this week, or it may be next week.
This entry was posted on Wednesday, January 16th, 2008 at 4:47 pm and is filed under Approximate Value, General, Statgeekery. You can follow any responses to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.