I’ve been a rabid fan of Baltimore sports since I went to my first baseball game in 1971.  The O’s lost 2 that day to the A’s (5/9/71) with Catfish Hunter and Vida Blue beating Mike Cuellar and Jim Palmer.  I didn’t know it at the time, of course, but it was probably the only time in my life I went to a doubleheader where all 4 pitchers won 20 games that season.  It was bat day, before society went to litigation hell, so we got our bats when we entered the ballpark, not by turning in a coupon later in the parking lot.  You can probably imagine that for a group of 2nd graders, a good bit of the fun that day came from making noise with those bats.  Since I know you’re interested, you can see the box scores and play-by-play results from that day at:

Bat Day May 9, 1971: Orioles v. A's

I’ve always enjoyed counting things in sports, and it’s never ceased to amaze me how often common conceptions I had or heard about baseball or football were just plain false when examined.  That got me into a number of fun hobbies.  When I got my first computer in 1984, my friend Dave Hollander and I built a database to track each at bat for that season.  We continued that project for about 4 years until the data became generally available in the late 1980’s.  After that, I worked with STATS inc in their early years doing pitch-by-pitch accounts from the press box until the mid 90’s.

Baseball and its fans have evolved tremendously since the 1980’s.  Bill James, Craig Wright, Pete Palmer, John Dewan, and others challenged the way we think about the game and provided both cogent hypotheses and the data to back them up.  Bill James once said that if Bobby Grich makes the Hall of Fame, it will be tremendous evidence that sabrmetrics has changed the historical view of the game.  I think there are bigger stakes for which we can still play. 

The greatest relievers are generally judged today based on saves and blown saves.  I think most followers of the game would agree that the extreme focus on saves and save percentage actually has guided the use of those relievers as almost exclusively 3-out specialists.  To me, the ultimate victory of sabrmetrics would be for relievers to be judged based on change in win probability.  So, for instance, if a reliever enters with no one out, the bases loaded and his team ahead by 1 run in the bottom of the 6^th inning, his team might have a 33% chance to win, but if he gets the side out without allowing a run, that chance might be 61%.  One might say he is responsible for a 28% shift in win probability.  This particular situation would be a very tough save situation, but a great opportunity to alter the win probability in his team’s favor, in this case even if a single run were allowed (and the save blown).  If change in win probability were to be generally understood, available, and reported on ESPN, I think relievers would be paid for it and the great ones would seek to be used for a high pressure set of outs 3 times a week.

Anyway, baseball has come a long way, but despite advances, I think football has a long way to go.

Why do you think it is that there aren’t many statistics on the back of football cards for offensive (most typically none) and defensive linemen?  I don’t know about you, but I’m tired of hearing unsubstantiated crapola about how Casey Hampton is better than Haloti Ngata.  From my point of view we can:

1.       Read each other’s subjective judgments, which we will never find particularly compelling

2.       Read the subjective judgments of “experts” that see each player play once a season but are happy to grade one of the above players an A- and the other a B+

3.       Compare defensive linemen based on tackles, sacks/QH’s, fumble recoveries, forced fumbles, and the occasional interception

4.       Count more things to improve the picture of performance

You can probably guess from the order that I choose number 4.  We’re not counting the things that matter, it’s not that the things that matter can’t be counted. 

So what do we need to count?  Well let me start by saying defense is tremendously complex and the interactions between players both on the same team and the opponent’s is a challenging modeling project to say the least. We’d have to start at its most basic level however, which are the players on the field each play.  Funny thing, though.  That’s not available currently from the NFL or publically available the same day from any source of which I am aware (Profootballfocus.com has this recorded for some 2007 games and will be expanding this season).  I knew I’d want to count other things, but 2 years ago I started with just this.

Participation-by-Play Analysis

At the heart of what I write about is Participation-by-Play (PBP) analysis.  For each Ravens defensive play of the game, I record every defender on the field along with the result of the play, number of pass rushers and some other data.  Once you get the knack of it, it takes about 3 hours to review just the defense and make notes.  With this information, we can look at a bunch of stuff.  For example:

·          Number of plays for each player (I’ll report this both as a gross number and % of the team’s “Real Snaps”)

·          A separate count of QB kneels, spikes, special teams plays where a punter or holder is credited with a run or pass (these plays are excluded from my defensive analysis although I enter the results to make sure the game totals match the NFL’s).

·          Yards per Carry (YPC) while each defender is in the game.  This will differ slightly from the YPC you’ll see in the paper, primarily due to the fact that kneels are not included.  Similarly, botched snaps will occasionally lead to a running or passing play on FG’s or punts, but neither of those is counted.

·          Yards per Pass Play (YPP) while each defender is in the game.  These specifically exclude spikes, where the defense is not involved (I told myself I would not repeat these exclusions in every article this year, so I’m going to get all this redundancy out of my system here).

·          Yards per Play All (YPPA).  This is the most significant statistic for a defensive player.

·          Sacks as a percentage of pass plays.  This is sacks/(pass attempts + sacks).  The denominator excludes both intentional spikes and plays on which the QB runs for positive yardage after dropping back.

·          Turnovers as a percentage of plays

·          Any of the above stats by the number of defensive backs in the game

·          Any of the above stats by the number of pass rushers

·          Any of the above stats by down or distance

·          Above statistics by offensive formation as defined below

·          Above statistics when the QB is in the shotgun or play action is executed

·          Above statistics by the number of pass blockers among eligible receivers (see below)

I use these statistics with my game observations in my weekly column “Filmstudy”

Other terms I use frequently when I write

·          LoS:  Line of Scrimmage

·          Standard:  4 DB’s on the field

·          Nickel:  5 DB’s on the field

·          Dime:  6 DB’s

·          Real Plays:  Plays that are not waved off by penalty, not kneels, not spikes, and not special teams plays

·          Pass rush expressed as X/Y.  X is the total number rushing the QB.  Y is the number of players that lined up on the LoS but dropped to coverage.  Y does not include anyone that is in press coverage against a standing receiver (although X might if, say a corner blitzes from press coverage).

·          OLS & ORS: Offensive Left Side and Offensive Right Side respectively.  Football labels its positions relative to their own side, so the Right Tackle lines up most frequently opposite the Left Defensive End.  That matchup is on the ORS.

·          Set blockers are counted on pass plays and include all eligible receivers kept in to block on the play.  They make no attempt to go out for a pass or they are counted as chip blockers.

·          Chip blockers (also counted only on pass plays) are those blockers who go out for a pass only after making contact with a pass rusher.

Information by Offensive Formation

There are a theoretically infinite number of offensive formations possible if you consider distance between players, distance from the LoS, etc.  Naturally, to analyze success of a defense versus any given offensive formation, it’s useful to categorize these into discrete groups.  There are several rules about formation that simplify:

·          The offensive team must have exactly 7 men on the LoS

·          The center 5 men on the LoS are ineligible receivers (a team will occasionally be penalized when a receiver is not set on the LoS outside each tackle).

·          One player accepts the snap from center

Here is an excellent Wikipedia link on both offensive and defensive formations:

Offensive and Defensive Formations

Anyway, all of the possible formations can be grouped by the number of wide receivers, number of tight ends, and the number of running backs (which always total 5).  Given the constraints and the definitions below, there are only 18 possible combinations.  Here is how I determine position:

·          A player is classified as a TE only if:

o         Within 1 body length of a lineman or another TE and in a 3-point stance; or

o         In motion where the motion ends within 3 yards of the LoS and within 1 body length of a lineman or another TE

·          A player is classified as a RB if:

o         More than 3 yards behind the LoS and between the TE’s

o         This includes players in motion meeting the above criteria

·          A player is classified as a WR if an eligible receiver and not in either of the above 2 classes

·          A player may be classified in any of the above 3 groups even if he does not report eligible

·          The player’s actual/roster position has no bearing on his position categorization for formation purposes

There are other folks who collect much more detailed positional data, such as Profootballfocus.com (a site I highly recommend, BTW), but I’ve chosen to simplify here for purposes of both speed of game review and having groups large enough to analyze the resulting data.  When I refer to formation, I will often refer to the formation as a 3-digit number, so if I say the Ravens faced the 500 formation 8 times and allowed just 2.8 YPPA, that means the Ravens faced a formation where all 5 eligible receivers were WR as defined above and the Ravens allowed 2.8 yards per play on those 8 plays.

A Closing Note on Model Simplicity

 I am really interested to hear your opinions on my methods and suggestions you may have for additional things to be counted.  I got a mess of very good suggestions with regard to the offensive line model columns I wrote last season.  I’ve got 2 basic rules, however:

1.        When I count something additional, I much prefer to count a raw, observable, statistic and not do subjective analysis on a play-by-play basis.  This sort of analysis is an enormous time drain and I find folks inherently don’t like models when they know subjective judgments (and their potential for bias) are at their underpinnings.

2.       The best things to count are discrete or may be sorted into discrete groups.  I can count the number of pass rushers, and it’s always a whole number.  I can group formations into 18 categories.  I can’t think of a way to fully describe pre-snap movement of a particular player that would be easily generalized or grouped.

Thanks to everyone for their encouraging ideas, posts, messages, and emails over the last 2 seasons.  This has been a mess of fun and let’s have a great 2008.  GO RAVENS!!!