# Wisconsin in numbers

Submitted by The Mathlete on October 1st, 2008 at 6:10 PM

As a numbers geek, I have spent a couple years trying to develop a system to go beyond the box score to analyze performance.  I have analyzed hundreds of box scores and play by plays to asign a value to each play based on down, distance and line of scrimmage.  Each combination of the three is given an expected value.

For instance, a 1st and 10 at the 20 will ultimately give you 1.49 pts each time. Obviously you are never going to score 1.49 point, but you could look at it as about 15 points (a little over 2 TDs) for evey 10 times the situation occurs.  Based on this, each play changes the expected points for the possession.  A turnover or failed 3rd or 4th down play results in the expected value dropping to 0.  In the case of a turnover, a penalty is added based on the resulting opponent field position in comparison to where the opponent would have started had there been an average punt.  A long interception with no return, little penalty, field position similar to a punt.  A long fumble/int return is a greated penalty because the offense not only loses their expected points, they give the opponent exceptional field position.

A couple notes on scores. Passing games see a lot more range in scores than running games. + is good for the offense, - is good for the defense.  Receivers have generally positive scores because only completions count towards their score.  Obviously Michigan's poor screen game against Wisconsin does not fit this trend.

Based purely on starting field position, if these two team were average the score should have been Wisconsin 41 Michigan 22.  Obviously a huge effort from the defense to hold Wisconsin to 25 pts and an a great second half for the offense, to put some drives together.

Without further ado, here are the scores for the week.

Offense Overall: -2.4 pts

Running Game: +1.3

Minor: 2 carries, +2.9

Threet:9 carries, +0.8

Brown: 1 carry, -0.7

Odoms: 1 carry, -0.9

McGuffie!: 15 carries: -1.5

Passing Game:  -5.2

Matthews: 4 catches, +4.3

Koger: 1 catch, +4.0

Minor: 1 catch, +0.6

Odoms: 2 catches, -0.2

McGuffie: 1 catch, -0.5

Brown: 3 catches, -1.3

Penalties (Offense): +1.5

For: 1 penalty, +1.9

Against: 1 penalty, -0.4

Defense Overall: -24.4

Run defense: -7.2

Other: 13 carries, +4.9

Evridge: 5 carries, -6.1

Hill: 22 carries, -6.0

Pass defense: -18.2

Gilreath: 5 catches, +6.0

Jefferson: 4 catches, -1.0

Penalties (defense): +1.0

For: 4 penalties, -2.3

Against: 2 penalties, +3.3

awesome. but shouldn't the passing game be 6.9, therefore making the offense overall 9.7?

The passing game is +6.9 on completions.  Incompletions and sacks bring the number down for the passing game, but aren't associated with any receivers, only Threet.

I understand evaluating the offense and defense based on what they should do with given field position. It is sort of like Value Over Replacement Player in baseball, where you're comparing the mean outcome to a given outcome. Very cool and insightful metric, IMO.

How did you come up for the expected point value per possession at a given field position? Is it based on all D-1A teams? If so, over what time period is the data taken? Did you account for time (i.e., drives that run out of time and, thus, don't reach the end zone deserve more of an "incomplete" than a "failure" on the part of the offense).

These numbers can go all sorts of places. Off the top of my head: evaluating non-conference schedules, ranking units nation wide, comparing likelihood of scoring for different conferences (does the SEC average more points when given the ball on the twenty?), etc.

The games used are from this season.  I included all teams from the 6 BCS conferences as well as a handful of "BCS busters" from the smaller conferences.  All games vs. D1A opponents were included (except for a handful that ESPN didn't have a Play by Play for).  So Mich vs Miami OH is included but Miami OH vs Temple is not.  Any drive that is ended by the half is excluded, as well as any drive that takes place when the game is well in hand is excluded.  I am working on adding a strength of opponent component that isn't currently in place.  Jacquizz Rodgers didn't have a high scoring game against USC last week when evaluated independently, but I am sure once the USC defense is factored, the value of his performance will increase dramatically.

Awesome.  I suppose the stregnth of schedule factor will have to emerge
from the point totals.  I'm looking forward to seeing how you work that part out, because it seems quite complicated.

I find this kind of analysis fascinating. I really like Ken Pomeroy's http://www.kenpom.com/ basketball system and how it has developed. I can see this evolving into the same type of system. Kudos to you for a great idea and getting the ball rolling to implement it.

Please keep posting and keep us up-to-date. Do you have the scores from the first few games? It would be interesting to be able to dis-aggregate the components of the ND game to explore objectively how good the game was outside of the turnovers.

I won't go into all the detail on the old games, but the projected score based on field position for the other games are:

vUtah: Utah 42 Mich 37

vMiami OH: Mich 22 MiOH 22

@ND: ND 34 Mich 25

The Notre Dame game is interesting because it was closer than I expected, but most of it was due to the fact that ND went into very safe, clock killing mode late and they got one score directly from the fumble return which is incremental to the numbers above.

The bottom line is the special teams and turnovers are killing the field position, Michigan has faced an uphill battle all 4 weeks.  2 of the the defense has been exceptional and the offense adequate, the other two didn't get the right combination to overcome.

Not sure if you guys are familiar with the Frimeau Efficiency Index:

This system rates college teams not on a game-by-game basis, as other computer rankings do, but on a drive-by-drive basis. For the NFL, Football Outsiders rates teams on a play-by-play basis, however given the number of college football teams and the difficulty of obtaining this data, it's understandable that drive-by-drive data represents the most disaggregated level of game information.

The system is a little kooky through 5 weeks - there's not that much information yet on each team, and it still includes a component of the questionable pre-season rankings. However, as the season wears on, I find this system to be the most accurate of all that I've seen - I was in "Bowl Pick'em" at my office last year and I cleaned up by leaning heavily on FEI.

Mathlete: great job. Not sure if you're familiar with FO or FEI, but there's a lot of great work going on with football stats nowadays.