Preseason Prediction: Michigan will end the year with a +8 Turnover Margin (TOM) or better (2011 was +7).  The prediction for TOM for M for this year is based on the prediction that M will be a very good team again this year and is not based on the actual TOM of last year. (Very good teams will have a TOM of +5 or better.)

Synopsis for Turnovers: M ended the game with a TOM of –1. That is 2 games in a row with a negative TOM while last year there were only 4 games with negative TOMs (SDSU, NW, Purdue, Iowa). Michigan has a TOM of –4 for the year (ranked #94). Denard was charged with another interception but it was clearly Vincent Smith's poor play that resulted in the tipped ball and subsequent interception. Even with the charged interception, Denard dramatically reduced his interception ratio from 11% in the first game to just 4% in this game. Air Force defied all odds for a triple option offense and ended the game with no lost fumbles (last year AFA was #107  nationally with 15 lost fumbles).

After 2 games, M has not fumbled once (ranked #1) but also has no interception takeaways (ranked #76).

 Synopsis for Expected Point (EP) Analysis: M lost the TO battle but won the game (approximately 17% of all college football games are won by the team that loses the TO battle). Without TOs, M would have won by approximately 6.84 points instead of 6 points.

Since Air Force scored a touchdown on the drive after the TO, why is the EP so small? This is where any analysis that relies on "what if" suppositions (as many analyzes do) may provide a less than satisfying result. It was 3-3 on the M44. If the pass had been incomplete, M would have punted on 4-3 at their own 44. Thus, M did not lose any EP because they would have still given up the ball on the next play. AFA took over on their own 44 instead of at their own 20 (assuming a net punt of 36 yards). Therefore AFA only gained the advantage of 24 yards of field position which results in just 0.84 EP. If the assumption is that Smith would have caught the pass for a first down then M would have lost an additional 2.35 EP for a total of 3.19 lost EP.

(See the Section on Gory Details below for how the adjustment for Expected Points (EP) is calculated.)

National Rankings: There is almost unanimous consensus that games against non-FBS opponents distort the statistics and muck up the analysis of data. Therefore, all rankings include games between two FBS teams ONLY and are from TeamRankings except for forced fumbles which is from CFBStats. The four columns with *** show the best correlation to offense and defense (per Advanced NFL stats).

The Gory Details

Details for Turnovers: Here is overall summary for all games by player (data in yellow was affected by this week's game).

Expected Point (EP) Analysis: Basically, the probability of scoring depends on the line of scrimmage for the offense. Therefore, the impact of a TO also depends on the yard line where the TO is lost and the yard line where the TO is gained. Each turnover may result in an immediate lost opportunity for the team committing the TO and a potential gain in field position by the opponent. Both of these components can vary dramatically based upon the down when the TO occurred, the yards the TO is returned, and whether the TO was a fumble or an interception.

Here are the details for the game.

The analysis is a bit tricky because: (A) the TO may directly result in lost EP for the offense but (B) only modifies the EP for the team gaining the TO because the team gaining the TO would have gotten another possession even without the TO (due to a punt, KO after a TD, KO after a field goal, etc.). The Net EP Gain must take into account the potential EP gain without the TO. The EP gain without the turnover is based on where the field position would have been for the next possession if the TO had not occurred.

The expected point calculations are based on data from Brian Fremeau at BCFToys (he also posts at Football Outsiders). Fremeau's data reflects all offensive possessions played in 2007-2011 FBS vs. FBS games. I "smoothed" the actual data.

Here is a summary of the smoothed expected points.