Mid-Week Metrics Hates Your Punt Comment Count

The Mathlete

[Ed-M: for the non-EV initiated, explanation of the maths are here. Also you can now use the tag "mid-week metrics" to read previous entries]

This week’s take on the week that was, and the week to come.

Charting Michigan’s odds of winning throughout Saturday’s game: Brief bumps around the two first half TD’s and then a big climb throughout the third quarter.

image

Best Three Plays:

1. Denard to Roundtree for 57 yards, 16%

2. Hawthorne intercepts Persa, 14%

3. Kovacs shuts the door on the fourth down option, 13%

Worst Three Plays:

1. Denard Interception #1, –13%

2. Denard Interception #2, –11%

3. Denard Interception #3, –10%

So the three first half picks were the three most damaging plays of the day for Michigan, and Denard only had one play in the top 3. At the end of the day he still ended up +65%!

For the season, Denard’s Win Percentage Added stands at +265%. Each team starts at 50% so they have added 50% per game times 6 games is 300% added. The defense, special teams and non-Denard offensive players have accounted for a total of 35%. Denard’s number is far and away the top in the country so far. Second is Virginia Tech QB Logan Thomas at +177%. Russell Wilson is next behind him at +158%.

Game Recap

Rush offense: –1, equal to ND for worst opponent adjusted performance of the season

Pass offense: +12, best of the season (Michigan’s best two games have both come despite 3 interceptions)

Rush defense: –3, worst performance of the season

Pass defense: +1, worst performance of the season

Special Teams: –2, 4th best of the season

Field Position: A rare win for Michigan, picking up a 23-22 win in expected points based on drive starts.

Players:

Denard was +12 (+10 passing, +2 rushing), second to ND on the season

Michael Shaw: +2

Vincent Smith: –1

Fitzgerald Toussaint: –5

 

(Expected value of offensives formations and more after the jump)

 

UFR + EV = Awesomely small sample sizes

I have linked up the UFR analysis with the expected value database to see what is working and what isn’t. Only offense now, hoping to have defense next week.

Formation Plays Avg Value Total
Shotgun trips TE 19 0.46 8.74
Shotgun trips 27 0.37 10.00
Shotgun 3-wide 93 0.32 29.90
Shotgun 2TE 18 0.25 4.52
I-Form 25 -0.12 -3.06
I-form Big 18 -0.24 -4.29

Shotgun good, I-form bad.

Defense Plays Avg Value Total
4-3 even 21 0.57 11.97
4-3 over 16 0.45 7.14
4-3 under 29 0.37 10.87
Base 4-3 43 0.30 13.04
Base 3-4 33 0.24 7.90
Nickel 17 0.23 3.96
3-3-5 stack 15 0.01 0.12
Base 4-4 11 -0.05 -0.51
3-3-5 under 15 -0.12 -1.84

A lot of this is based on opponent, but Michigan has done very well versus the 4-3.

Play Call Plays Avg Value Total
Zone read keeper 11 0.54 5.93
Zone read dive 37 0.34 12.53
QB power 19 0.28 5.35
Hitch 10 0.06 0.60
Power off tackle 20 -0.23 -4.52

Zone Read FTW, literally.

RB TE WR Plays Avg Value Total
1 0 4 26 0.43 11.14
1 1 3 126 0.37 47.05
1 2 2 48 0.31 14.74
2 1 2 39 0.10 3.91
2 2 1 21 -0.09 -1.95

2 RBs is bad news.

Gutless Punt Decision of the Week

Trying out a new feature this week in honor of Brian’s twitter callout of Ron Zook’s punt from Indiana’s 30.

Each week I will pick the worst punt decision of the week. The call will be based on field position, yards to go and game situation and ultimately be based on my gut looking at all the relative kicks.

In this week’s slate of games, there were 93 punts from opponent territory. Of these 12 came with the game within two scores and 3 or less yards to go. The ultimate chicken of the week goes to Frank Solich at Ohio (NTO). Facing a 4th and 1 from the Buffalo 36, Solich decides to “trust his defense” and punt the ball away. The gutless decision of the week is based on situation and not result, but man does the result really make this one good. Ohio punter Paul Hershey boots the ball 11 yards and Buffalo proceeds to march 75 yards for a touchdown on the ensuing drive. Ohio lost by 1. [Ed-M: For context, Buffalo's rush defense is among, if not the worst in the country.]

Updated Season Projection

image

About an 80% chance of hitting double digit regular season wins. With the expected win versus Northwestern, odds of an undefeated regular season stay at 1 in 11.

Remaining games with change vs last week’s odds:

@Michigan St: 47% (-1%)

Purdue: 100% (nc)

@Iowa: 70% (+7%)

@Illinois: 40% (-5%)

Nebraska: 77% (+3%)

Ohio: 93% (-2%)

 

Rankings

My top 5

1. LSU

2. Oklahoma St

3. Boise St

4. Wisconsin

5. Oklahoma

Other Big Ten:

8. Illinois

10. Michigan

19. Michigan St

27. Penn St

32. Nebraska

64. Iowa

66. Purdue

67. Northwestern

94. Ohio

105. Indiana

107. Minnesota

 

Michigan St Preview

Rush Offense:

Michigan: +6, 5th (Oregon), 1st

Michigan St: +4, 6th, 1st

Denard: +4, 2nd QB rusher nationally (Klein, Kansas St)

Fitz: +1, 72nd RB (Smith, GT)

Vincent Smith: +1, (dnq)

Pass Offense:

Michigan: +3, 28th (Baylor), 3rd (Wisconsin)

Michigan St: +5, 12th, 2nd

Denard: +3, 51st passer (Griffin, Baylor)

Junior Hemingway: +5, 63rd (Kendall Wright, Baylor)

Rush Defense:

Michigan: +3, 23rd (LSU), 5th

Michigan St: –2, 98th, 11th

Le’Veon Bell: +1, 46th

Edwin Baker: –1, 188 (of 217)

Pass Defense:

Michigan: +1, 50th (UCF), 7th (Penn St)

Michigan St: +2, 42nd, 5th

Kirk Cousins: +3, 46th

BJ Cunningham: +8, 16th

Special Teams:

Michigan: –1, 90th (Texas Tech), 12th (Purdue)

Michigan St: +1, 57th, 6th

Prediction:

I’ll go straight numbers here and call it 28-27 Michigan St but this game is a toss-up. Every unit is about comparable to the oppositions with the exception of Michigan’s run defense vs Sparty’s ground game in Michigan’s favor. A slight Michigan advantage there is offset by the game in East Lansing. We find any Hoke magic and Paul is heading back to Ann Arbor.

Comments

LBSS

October 12th, 2011 at 10:06 AM ^

+1

In particular, an explanation of what, e.g., "Michigan: +3, 28th (Baylor), 3rd (Wisconsin)" means would be awesome. My puny brain can make neither head nor tail of that gibber-jabber.

Or a link to the first place you used that stuff.

Thanks.

joeyb

October 12th, 2011 at 10:53 AM ^

IIRC, that number is in PAN or Points Above Normal. +3 means that unit will earn our team a 3 point advantage more than the average team. Basically, if MSU is the average team, +3 means that we can expect a 3 point win. So, he calculates the PAN for all of the units and gets a general idea of what the spread would be. He also gives a 3 point favor to the home team. Each point is work 3% in probability. So, he starts at 50% and a 3 point favor puts the home team at 59%. Then he adds or subtracts 3% from that for each PAN and comes up with a percentage. He'd have to confirm this, but I'm pretty sure most of that is correct.

justingoblue

October 12th, 2011 at 11:04 AM ^

Zero PAN means you are completely average. For a BCS conference team like Michigan this typically means bottom third of the league. A three-points swing in PAN typically equates to an additional win or loss over the course of a season.

+7 will put you around the Top 25 on the season

+14 is typically Top Ten and potential BCS game

+21 is best in class and probably playing for a national championship

The top rated team I have is Florida 2008. They finished +13 on offense, +7 on defense and +3 in special teams. The top Big Ten team is Ohio 2005 at +19 (7/9/3). The top Michigan team is 2006 at +14 (4/6/4). They come in at 50th overall in the last 8 seasons.

Logan88

October 12th, 2011 at 9:59 AM ^

I'm trying to figure out how Illinois is better than UM. Both teams are 6-0, both have good offenses and good-ish defenses, both have pounded a few bad teams, both have squeaked by the one good team they have played but against the two opponents we have in common UM has won handily (24 vs. WMU and 18 vs. NW) while Illiniois has squeaked by both.

BlueBulls

October 12th, 2011 at 10:53 AM ^

I'm loving the Mid-Week Metrics posts, but one thought regarding the best and worst plays:

Would it be possible to note the play number of each instead of (or in addition to) just ranking them 1 2 3?

I'm pretty sure I can find them in the chart, but it's still tought to tell since we don't know what play number they were. It would also be interesting to how the Win Percentage responds to the team's play calling after important plays.

thisiscmd

October 12th, 2011 at 11:11 AM ^

How long does this take you Mathlete? This requires a ridiculous amount of data. There must be a streamlined way of accessing the data and crunching your numbers. Or you just don't sleep.

Impressed as always...

jamiemac

October 12th, 2011 at 12:43 PM ^

The real questions are.....

Will KState keep going undefeated (The JCB won on them last week!!!)

And, natch, will Northwestern cover the +6.5/7 at Iowa this week

I'll hit submit, wait and read.

Thanks, Mathlete

Eye of the Tiger

October 12th, 2011 at 2:39 PM ^

I remember you predicting us winning 8 or 9 (I think), and obviously that was off a bit, but that's just N=1 and I'm not asking you to be Nostradamus here...I know that even good predictive models are only accurate 50+% of the time.

But I would like to know how well you predicted other Big 10 teams at the midseason point last year, so I can decide just how much Kool-Aid I should drink in advance of Saturday's game.  Thanks!

joeyb

October 13th, 2011 at 6:03 PM ^

Just a thought for a future, off-season diary...

It would be cool for you to do preseason predictions on teams based on player PAN. Basically, if a player is +12 PAN for their team and they leave, replace them with a calculated average for a new starter in that position. Maybe the new starter has some playing time and already has a +2 PAN and you can extrapolate that. The other option would be to assign value to players based on recruiting ranking or some combination of the two (5* players with +2 PAN after their sophomore year average +5 PAN as a junior in their senior year).

Using this, you could estimate a team's PAN and do preseason rankings, estimate EOY records, etc. This is something that I've thought about for a while, but I don't have the means of doing it. You, on the other hand, already have all of the data and most of the formula done.