informative post. Thanks for the numbers and I hope that it all falls into place.
spoiler alert: i linked this
Synopsis: After 6 games, Michigan is currently ranked #14 in scoring offense and #75 in scoring defense. Based on these rankings, M has a 43% chance for a +5 WLM (9-4 or better) season and an 83% chance for a winning season. The offense definitely had its worse game of the season scoring only 17 points (primarily due to TOs). Rushing yardage was about 50% of the average for the year whereas passing yards were fairly close to the average. The defense continued to struggle allowing 9 more points than average and about twice the average rushing yardage. Defense passing yardage was slightly below the average. Michigan's current PPG is 26.8 so the defense must hold teams below this number to improve their performance.
I always use scoring stats because yardage stats are inherently flawed. Being #75 in scoring defense is not good but U-Ms defense is not as bad as the #112 in total defense indicates. According to the S&P+ rankings at Football Outsiders, Michigan is ranked #62 in total defense.
Based on the FEI (Fremeau Efficiency Index), Michigan is predicted to win between
7.5 and 8.1 8.1 and 8.5 (ED: corrected) games (excluding bowl game but adjusted with +1 for U-M's one FCS opponent).
Based on the FEI, Iowa is favored by just a single point with just a 51% PWE (projected win expectation). Using the Sagarin Predictor, Iowa is favored by 3.4 points (note the Sagarin Elo-Chess actually has M as the favorite by 6.7 points and the Sagarin overall rating has M favored by 1.4 points). (Vegas Odds Opened with Iowa favored by 3.0).
I am surprised at how close this game is predicted. The SoS adjustments are especially interesting because the FEI SoS Algorithm (explained here) shows Iowa with a more difficult schedule than Michigan but Sagarin has just the opposite. Iowa has also had an extra week to prepare due to their bye last week. IMO, Michigan will have to play its best game of the year and end up with a positive turnover margin to win.
Overall this year, U-M is averaging 3.2 points per possession (PPP) and 46 YPP. The defense is giving up 2.2 PPP and 38 YPP. With an average of 12 possessions per game for each team, this translates into a 12 point advantage for Michigan.
DETAILS: Here are the FEI numbers ( FEI Forecasts and Football Outsiders FEI ). FEI is a weighted and opponent adjusted season efficiency and is expressed as a percentage as compared with an average FBS team.
Note that FEI completely excludes all non-FBS data (the W-L record is only for FBS games, etc.). Therefore, you need to add 1 to the projected numbers for FBS-MW to get the final predicted wins for U-M this year. Or, if using the FBS-RMW, add 1 to the current win-loss record to get the final predicted wins for the year.
The FEI is a drive based analysis considering each of the nearly 20,000 drives each year in college football. The data is filtered to eliminate garbage time (at the half or end of game) and is adjusted for opponent. A team is rewarded for playing well against good teams (win or lose) and is punished more severely for playing poorly against bad teams than it is rewarded for playing well against bad teams. I've included the GE basic data so you can see the impact of adjusting for opponent. (See: Football Outsiders Our Basic College Stats )
Here are the Sagarin Ratings.
Sagarin uses two basic ratings: PREDICTOR (in which the score MARGIN is the only thing that matters) and ELO-CHESS (in which winning and losing only matters, the score margin is of no consequence). The overall rating is a synthesis of the two diametrical opposites, ELO-CHESS and PREDICTOR.
Per Sagarin: ELO-CHESS is “very politically correct. However, it is less accurate in its predictions for upcoming games than is PREDICTOR”.
Here is the U-M vs. Iowa National Statistical Rankings with the advantage for each category indicated (all categories within 10% are considered a "push").
Here are the week by week National Statistical Rankings for Michigan (cumulative thru the week indicated):
I have included the major rankings for offense and defense but scoring rankings show the best correlation to winning and losing. Scoring rankings are based on PPG. Rushing, Passing, and Total rankings are based on YPG.
Here is the basic data for Michigan (each individual week followed by totals and then average per game). I've included Total Possessions for Offense & Defense along with the calculated data per possession. Number of possessions do not include running out the clock at the half or end of game. Offense Plays and Defense Plays are better indicators than Time of Possession.
Using Scoring Offense and Scoring Defense National Rankings for the past 5 years (FBS AQ teams only), this table shows the percentage of teams that finish the season with a +WLM and a +5 WLM. For example, teams that finished in the Top 40 in both offense and defense had a 100% chance to be +WLM and an 82% chance to be +5 WLM (9-4 or better).
Each year, of the 66 FBS AQ teams, 65% (43 teams) end up with a + WLM and 36% (24 teams) end up with a +5 WLM.
informative post. Thanks for the numbers and I hope that it all falls into place.
Thanks for this analysis. I still don't understand 90% of this information but the portions I do understand look very promising. The bottom line is that the offense is averaging one more point per possession on offense than defense which correlates to wins if the averages play out in reality. Here's hoping at least that occurs on Saturday!
a whole lot more if I knew how to read charts like you! Impressive stuff... interesting that the per possession numbers on the ND game look very similar to those against MSU (except reversed Pass/Rush), but the outcome is so different. That must be turnovers?
Agreed - I thought the MSU game a whole lot worse than ND, but the post earlier about Jonathan Chait's reexamination of the game showed that it wasn't as bad as I thought...thanks to this post and others for helping me maintain perspective...
Hard to believe this defense ranks in the top half of FBS by any measure. I guess strength of schedule is what turns 112th/75th (total defense based on yardage/scoring) into that very average number, but opposing offenses (based on S&P+) haven't necessarily been the toughest:
And the remaining schedule:
Iowa: 4th (!!)
Illinois: 14th (!!)
The homestretch almost looks even more threatening with those numbers...
Next week FEI adds the offensive and deffensive efficiency ratings for the season (they are not included earlier due to limited data). It will be interesting to see what these indicate. But, I do expect the FEI O&D ratings to be a lot different than S&P+. I will be large sums of money that M is closer to #60 than to #112.
I have been using the S&P+ because that is one that is available early in the season. But, if you compare FEI overall rankings versus S&P+ overall rankings it is pretty obvious that the S&P+ algorithms are highly volatile and very different than most other systems. (For example, S&P+ has Illinois ranked #10 -- ahead of LSU, MSU, Missouri, Oklahoma, etc.)
FEI has Illinois ranked #40.
I enjoy these posts, but I often feel I leave them with an overwhelming headache and incurable anxiety. But, I still subject myself to them, and will continue to, because hey--I don't have anything better to do.
"Michigan will have to play its best game of the season." This team is capable, but PREDICTOR says, "Iowa by a field goal."
Thanks for taking the time to put this together. I've always enjoyed statistics, and even create my own end of the season computer ranking ever since I did a quick statistical calculation that predicted the USC-Texas National Championship game to within half of a point.
However, I'm still confused by the RMW. Since the explanation given on the FO site (FBS RMW: Mean Wins expected for only the remaining games on the given team's schedule.) explicitly says it only considers the remaining games, why is the +1 required? I asked this last week, and I still don't quite understand.
For example, take Nevada (28th on the FO FEI ranking list). They're listed at 5-0, but are actually 6-0 with a win over Eastern Washington. Their FBS RMW is listed as 5.6. If we add 1 to this, we get 6.6, giving them a total season of 12.6 (they play at Hawaii, making a 13-0 regular season possible). This means, even assuming they're 100% likely to win every game, that they are 60% likely to beat Boise State (home game, but still), when BSU is #2 in their rankings.
However, at the same time, BSU has an FBS RMW of 6.6 with 7 games left on their schedule, which is the same scenario as Nevada (100% in every game except Nevada, where they are 60% favored). Since both teams can't be 60% favored, I'm incredibly confused. Thoughts?
(As an aside, there are 34 remaining games in the Big Ten, with Illinois and Indiana each having 1 additional non-conference game left to play, Fresno State and Arkansas State, respectively . If we total up the RMW for all Big Ten teams, we get 35.3.)
Oh, now I see. You can't add 1 to the current record and then add another 1 to the FBS RMW unless a team has an FCS game remaining.
For Nevada, their record is listed as 5-0 but is really 6-0 with the FCS win. 6-0 + 5.6 (FBS RMW) = 11.6 instead of the 12.6 you indicated.
Or, if you take Nevada's FBS MW of 10.0 and add their FCS win (+1) you end up with a total of 11. Giving a projected total wins of 11 to 11.6
For Michigan. FEI lists our record as 4-1 but it is really 5-1. We do not have any remaining FCS games so just add the FBS RMW. 5 + 3.5 = 8.5
Or, you can take the FBS MW of 7.1 and add the +1 FCS win to get 8.1.
Thus, M has a predicted total of 8.1 to 8.5 wins.
So, I got that wrong in the OP -- it is now corrected.
I think the problem with looking at the RMW or MW for multiple teams is basic statistics. FEI defines RMW as the "average number of remaining games a team with the given FEI rating would be expected to win against the given schedule".
Some teams will perform remaining games above average, some below average, and some probably about average.
FEI uses non-linear groupings to determine PWE (projected win expectation) so I don't think it is possilbe to compare and/or combine multiple teams. See the discussion on Projecting team records here:
I think, since we're not strictly considering specific games, it will work. Since (and I'm assuming here) the RMW is just a sum of the probabilities of winning each matchup given the respective FEI ratings of each team, the net total of all the probabilities should equal 34 (depending on the rounding used) for all the Big Ten teams, if only Big Ten games are considered.
If we use the OSU-USC numbers in the table, and add them up:
OSU: Projected 9.562 total wins (10-1, ignoring a FCS game), 7.106 in conference (7-1)
USC: Projected 9.783 total wins (10-2, no FCS game to ignore), 7.635 in conference (8-1)
However, OSU's non-conference predicts 2.456 wins, which rounds to 2-1. So there's somewhat of a breakdown there. But I think if you consider an entire remaining schedule, the probabilities should work out. In other words, when Michigan and OSU play at the end of the season, the RMW column should be something like 0.2 for Michigan and 0.8 for OSU.
As for the Nevada-Boise matchup I discussed, yeah, the 12.6 I said was "incorrect". However, I was using it to show why I didn't think RMW needed a +1. Of course, if you're just adding RMW to the record to get an end of the season projection, you do need the +1 to account for the FCS win. I think it'd be better stated as something like "add +1 to the current record and the MW data to account for wins against non-FBS teams".
The spread based on FEI is Iowa by a single point with a 51.0% Predicted Win Expectation.
I continue to be amazed this game is predicted to be so close.
I've updated the OP.
Me too. I had a dream (or nightmare) that the score was 22-70, Iowa.