Tuesday, November 18, 2008

A Statistical Explanation for the Prevalence of Fall Coronaries in Gwinnett County

As Auburn was driving for the potential winning score on Saturday, I was thinking to myself "isn't this pretty much the same as every Georgia game this year?" With the exception of the two games against elite opponents (Alabama and Florida) and the one game against a really bad opponent (Tennessee?!?), every game has been the same: relatively low-scoring and ending with the opponent trying to drive for a tying or winning score in the final minutes. South Carolina, Vandy, Kentucky, Auburn, all the same. The LSU game is the only exception and even there, LSU was driving late, albeit in an attempt to close the game down to one score.

Four close games, all against teams outside of the top 30. Any good team is entitled to a close game or two against an inferior opponent. Also, in most years, I'm willing to cut SEC teams a little extra slack in playing close games because of the depth of the conference (especially defensively). That said, Georgia seems to have played a whole bunch of close games this year. And this year is hardly unique. Richt's first SEC Champion started 8-0 and half the wins were by one score. Richt's 2004 team, the one closest this this year's squad in terms of expectations, won half of its ten games by one score. The 2007 Dawgs seemed to play every other game close in the first half of the year before catching fire in the second half.

Anyway, there are two possibilities here. The first is that Georgia doesn't play any more close games than any other major power and I'm experiencing a sensation that affects most college football fans. The second is that Georgia does play more than its share of close games. In honor of Nigel Tufnel, I decided to look at the top 11 BCS Conference teams (as measured by winning percentage) since Richt's first season in Athens and rank them by the number of games decided by one score in which they played. To the numbers we go!

  1. Michigan - 43 games - 23-20 - 43% of all games
  2. Auburn - 40 games - 27-13 - 40% of all games
  3. Georgia - 40 games - 26-14 - 39% of all games
  4. Miami - 35 games - 20-15 - 36% of all games
  5. Florida - 36 games - 19-17 - 36% of all games
  6. Ohio State - 33 games - 22-11 - 33% of all games
  7. LSU - 33 games - 24-9 - 32% of all games
  8. USC - 31 games - 17-14 - 31% of all games
  9. Virginia Tech - 29 games - 12-17 - 28% of all games
  10. Texas - 28 games - 20-8 - 28 % of all games
  11. Oklahoma - 23 games - 14-9 - 22% of all games

So what do we learn from these numbers?

1. One of the basic precepts of the Sabrmetric revolution in baseball is the conclusion that a team's record in one-run games is mostly random. A team that has a good record by virtue of winning a disproportionate share of close games is likely to regress to the mean as the season progresses. The best way to measure a team is to look at its scoring differential rather than its record. It appears that this is not the case in college football, as the top programs win roughly 60% of their close games. It's possible that top teams are likely to be leading and give up late scores to make games look closer than they appear, but that explanation doesn't ring true to me. The best explanation I can offer is this: if Georgia and Vandy are tied at 17 with five minutes to go, Georgia still has a better team and is more likely to get the winning score. The question then becomes this: why was Georgia tied with Vandy after 55 minutes to begin with?

2. It is not at all surprising to see Michigan at the top of this list. There are a number of reasons why Michigan fans were not overly upset to see Lloyd Carr retire despite the fact that he had a good record in Ann Arbor. One of the most prominent reasons was Lloyd's tendency to make each game a self-fulfilling prophecy. Lloyd assumed that most games would be decided in the fourth quarter. He perpetually underrated the talent advantage that his teams had over most of the team's on their schedules. (Lloyd does deserve credit for recruiting well and creating the talent advantages in the first place.) Thus, Michigan would employ sub-optimal offensive strategies (read: running between the tackles and horizontal passing) until threatened, at which point Michigan would use its best assets: NFL-caliber quarterbacks, wide receivers, and offensive linemen in a spread passing attack. If Lloyd would have used the Citrus Bowl offense that ripped Florida to shreds as the team's base offense, then the Appalachian State debacle never would have happened. If he would have used the spread passing attack with Tom Brady as the base offense in 1999 instead of giving the ball to Anthony Thomas 25 times per game from the offset I, then Michigan wouldn't have played nine games decided by one score (nine!?!) with a future NFL Hall of Famer under center.

3. I was expecting to see Ohio State higher on this list, but after the Bucks played 20 one-score games in Jim Tressel's first three years, they've only played 13 in the almost five seasons since. The are a couple lessons to be drawn here. One is that Ohio State has played in a lot fewer close games since they went to a spread passing game with Troy Smith and now Terrelle Pryor. One could also argue that Tressel has accumulated more talented teams now that he's playing with his own recruits and/or that Tressel's coaching style has evolved and he's realized that it's better to dominate opponents than it is to assume that every game is going to come down to one or two plays. The second lesson is that whatever magical skills Tressel possessed in 2002 and 2003 to win close games were a mirage. Tressel was 15-5 in one-score games through 2003 and is 7-6 in one-score games since that time.

4. I've been a Mack Brown defender since his early years at Texas when he got the mindlessly applied "can't win the big one" label, but even I was surprised to see that Texas has the best winning percentage in one-score games in this cohort. That runs counter to the notion that Mack assembles talented teams and just rolls the ball out so they can play. He can clearly make good tactical decisions when his teams are pushed. Conversely, Frank Beamer is seen as a coach who gets the most out of more limited talent, but Virginia Tech has the worst record of the teams in the cohort and it isn't close.

5. I was interested to compare Ron Zook and Urban Meyer in this department. Meyer has coached 16 one-score games out of 49 total games at Florida (.326). Zook coached 18 one-score games out of 38 (.473).

6. So what does this all mean for Mark Richt, other than confirming my sense that his teams are more likely to play tight games? I might be over-interpreting here, but Richt doesn't seem to get as much out of his offensive talent as he should, which prevents Georgia from dominating its opponents. Richt developed his offensive approach at Florida State in conditions where there was no pressure to maximize production using scheme and play calling because of the Noles' huge talent advantage. Also, I think it's fair to say that playing a lot of close games has a negative correlation with winning a national title. The three teams at the top of the list have not won national titles in the time period covered; six of the bottom eight teams on the list have. (A counter point: two of the three teams that have played the lowest percentage of close games have not won a national title since 2001. Another counter-point: Auburn hasn't won a national title, but they did have a perfect season during this time period.) It's not impossible to win a national title by playing a style that leads to a lot of close games, but it sure does increase the degree of difficulty.


hoodawg said...

Michael, I'm willing to buy your theory on Richt's offensive underperformance (particularly in years prior to 2008). But wouldn't we need to run a separate comparison to see what the average total points scored in those one-score games are, per team, to know whether it's the offenses or the defenses that are lacking? You could have a team that routinely wins 45-38, and the failing would be defensive, not offensive.

I expect Georgia would come out with a lot more low-scoring games, particularly early in the season, but it's just a guess.

Anonymous said...

Re: Michigan, the entire offense was healthy for the whole Citrus Bowl, unlike Appalachian State and Oregon at the start of the season. So the spread passing attack wouldn't work so well without a healthy Henne, Hart, etc. Plus, Henne was brutal in the first two games of the year; a spread passing attack would have only increased his number of picks. (By mid-season, however, Henne was so good that he single-handedly won road games at Illinois and Michigan State with his right shoulder falling off. The shoulder then did fall off against Wisconsin and Ohio State before he got healthy for UF in the bowl game).

Michael said...

Hoo Dawg, that's an interesting idea. I'll need to do a follow-up post on that subject.

Anon, I was prepared for that criticism, which is why I limited the line about 2007 Michigan to the App. State game. The offense was healthy in that game, but it was unproductive because Michigan made itself easy to defend by being very conventional: zone running and fly patterns. Michigan had the talent to go four-wide out of the shotgun. App. State would have had no prayer of defending against that sort of offense. (Does a I-AA team really have nickel and dime corners who could cover Michigan's receivers?)

As for the end of the season, I agree with you. Henne and Hart were injured for the last two games, which meant that any offensive system was doomed to failure.

Anonymous said...

I disagree with conclusion #1: I think its more likely that you just have a small sample size, which allows the effect of the one-score games to determine the top-teams (iow, teams that get disproportionately lucky and win an inordinate number of one-score games will have the best overall records). After all, your theoretical rationale for why Georgia would be likely to defeat Vanderbilt in a tie game with 5 minutes left....is equally persuasive as a reason that the 2003 Tigers (worst team ever) should have lost most of their one run games. In a close game in the 8th inning, one team has the 2003 Tigers' lineup and bullpen (or heavens forbid, starter) that won 19% of the games decided by more than one run (24-101), and one has *any other major league lineup lineup and bullpen.* Yet, astoundingly, the 2003 Tigers were 19-18 in one run games! The second best record in one run games in 2010 belonged to....the Baltimore Orioles! 29-21 in one run games despite a lousy lineup and terrible bullpen, 37-75 in the rest of their games (33%).

I mean, was Les Miles anything other than lucky in his close games in 2010? Can you attribute one iota of coaching skill or "reserve capacity" to his wins against Tennessee, WVU, Alabama or UNC? Its to his credit that LSU was good enough to play a great Alabama team closely, but come on; that was a fluke. Similarly, its not like Carr had anything to do with the abhorrent officiating at the end of the 1999 ND game, or Wisconsin and Illinois dropping punts inside their 20 yard lines, or MSU's kicker missing multiple chip shots, etc.

Close games are luck.