Possibly the closest Super Bowl matchup in history also poses the question: how much does it mean when certain aspects of an NFL team improve dramatically in the second half of the season?
10 Oct 2008
by Bill Connelly
Last week, the conversation topic on Varsity Numbers was WinCorr -- specifically, the correlation of certain Varsity Numbers-centric statistics to wins and losses. This week will focus first on what happens when the WinCorr idea is applied to specific teams, and second on a couple of comments and suggestions from last week.
The idea of a team-specific Win Correlation is simultaneously interesting and limited. As you'll see, it provides a unique footprint for each team in the country; it shows you the areas that tended to decide each team's games. At the same time, though, it brings back the sample size question: If you are looking at performance on a per-game basis, that means you will end up with a sample size between 12 and 14, which is obviously quite low from a statistical standpoint. Unlike a lot of statistics, though, Team WinCorr offers an explanatory viewpoint instead of an evaluative one, and that makes the sample size issue more palatable. Team WinCorr is meant to complement what you already know about a team, rather than tell you something you did not.
To illustrate how varied these blueprints can be from one team to another, here are a couple of examples (and once again, these are Spearman correlations, comparing a given category to the percentage of points scored by the team in question for each game) of two teams playing a pretty big game this weekend:
|LSU, 2007 WinCorr|
|Defensive Passing EqPts||0.866|
|Defensive Passing S&P||0.861|
|Defensive Passing PPP||0.858|
|Defensive Passing Success Rates||0.831|
|Defensive Success Rate, Passing Downs||0.827|
|Defensive Passing S&P+||0.826|
|Defensive S&P+, Passing Downs||0.823|
|Defensive PPP, close games||0.820|
|Defensive EqPts, Passing Downs||0.807|
As was mentioned last week, it is hard to draw too many conclusions from these correlations, being that the numbers are so close, the correlations are so high, and again, the sample size is small. However, this does suggest that, relatively speaking, LSU's offensive performance did not make a major impact in the results of their games. The Bayou Bengals were solid offensively, but their margin of victory (and their two defeats) were often decided by defensive breakdowns.
Why defensive breakdowns? Seven of these top 15 categories had to do with passing defense despite the fact that LSU was solid in the secondary (No. 23 in pass efficiency); plus, three of these categories were related to passing downs, suggesting LSU had the occasional breakdown on passing downs (against, predictably, Arkansas and Kentucky in particular), and when they did, their odds of losing naturally went up.
For the record, here were the top 3 offensive categories:
1. PPP, first downs (0.677)
2. EqPts, first downs (0.655)
3. S&P, first downs (0.639)
The Nos. 5 and 6 offensive categories were also related to first downs. This suggests that the LSU offense's job was, at its most simplified, to move the ball well enough to win the field position battle and spell the defense. If they got some yards on first downs (and therefore made themselves more likely to move the chains and eat up the clock a bit), they were likely going to win.
In a way, this list was a good sign for LSU heading into 2008. While they had to replace a few defensive cogs -- defensive tackle Glenn Dorsey and safety Craig Steltz in particular -- the defensive depth was good, and the biggest question mark was at quarterback. To maintain their relative success level, however, they just needed to find a quarterback who was a competent game-manager, not a game-changer. In the first half against Auburn, redshirt freshman quarterback Jarrett Lee almost changed the game in a negative way, throwing a terrible telegraphed interception that was return for a touchdown. However, when Andrew Hatch got knocked out of the game later on, Lee settled down and performed just well enough to allow other play-makers (running back Charles Scott, wide receiver Brandon LaFell, linebacker Kelvin Sheppard, cornerback Chris Hawkins) to win the game. Lee (and/or Hatch) will have another opportunity to not hurt his team this weekend in a big game against Florida in Gainesville.
|Florida, 2007 WinCorr|
|Offensive Rushing Success Rate, red zone||0.899|
|Offensive S&P+, close games||0.856|
|Offensive S&P, close games||0.849|
|Offensive Success Rates||0.847|
|Offensive Passing Success Rates, close games||0.845|
|Offensive PPP, close games||0.843|
|Offensive Passing S&P, close games||0.837|
|Defensive Rushing S&P+, non-passing downs||0.832|
|Defensive S&P, pressure situations||0.827|
|Offensive Passing S&P||0.827|
|Offensive Points Per Play||0.825|
|Offensive Success Rates, close games||0.825|
|Offensive S&P, red zone||0.822|
|Offensive Passing S&P, first quarter||0.818|
As good as Florida's offense was last year, you would almost expect that the Gators' list would be littered with defensive categories. But what Team WinCorr shows you is a list of categories that varied. If a team was consistently good or bad in a given area, that area wouldn't show up on this list. And while the Gator offense was great in '07 (No. 1 in the S&P+ rankings), it didn't always come through in clutch situations, and if there was a chink in the Tebow armor, it was efficiency (success rates) instead of explosiveness (Points Per Play). Now, all of the correlations on this list are quite high, so saying one thing was more important than another is a marginal argument, but Florida's failures were few and far between in 2007, so it still bears mentioning.
Being that Florida was breaking in a lot of young players on defense, and they came into 2008 with a year of pressure situations under their belt, the obvious conclusion was that the defense would respond better to pressure than it did last year, and the offense wouldn't have to do as much to keep the Gators in the title hunt. Phil Steele even picked them No. 1 because of their great offense and defensive improvement. However, while the defense has indeed improved so far in 2008, the offense's slight regression has held them back. After putting up more than 40 points 10 times in 13 games last season, Florida has managed the feat only once in five games in 2008. Was the presence of all these high correlations for offensive numbers a sign of foreboding for the Florida offense? It is impossible to say what role these numbers can play yet, but it is something to watch once a few years of data are in place.
Here are some responses to last week's comments.
I really don't like how you switch standards from percentage of points to pure wins on the + stats just because you got higher correlations that way, for 2 reasons. Higher correlations are good, but by changing the standard you have made your results incomparable to each other.
I also really, really think that you should run win correlations on games just between teams in the Top 25, too. It is a major, major assumption that games between teams in the Top 25 are in any way similar to games between teams in the Top 25 and teams at the bottom of the league.
It's entirely conceivable that a team can be easily built to dominate inferior opponents, but poorly constructed to match up with teams of equivalent strength.
The database wasn't set up with Top 25 rankings in mind, but here's a list that attempts to get at this sentiment:
|Win Correlations, conference games only|
|Defensive S&P+, close games||0.653|
|Offensive S&P+, close games||0.635|
|Defensive Passing S&P+, close games||0.581|
|Defensive Passing S&P+||0.579|
|Total first down rushes||0.564|
|Offensive Passing S&P+||0.560|
|Offensive Passing S&P+, close games||0.551|
|Defensive S&P+, non-passing downs||0.534|
|Offensive S&P+, non-passing downs||0.523|
|Points Per Play||0.522|
|Rushing EqPts, first downs||0.507|
|These are again Spearman correlations between
given categories and percentage of points.
In theory, while you can indeed build a schedule around inferior opponents, you cannot avoid conference opponents. There are plenty of bad conference matchups -- Oklahoma/Baylor, Boise State/Utah State, anybody in the Big East/Syracuse -- but conference matchups are more competitive, on average. So this should give you a good idea of what is more important in competitive games. This list shows that the straight-forward stats come out with the highest correlations to wins. And more interestingly, three categories actually had higher correlations than points scored.
And for the record, Success Rates had only a 0.370 correlation, 0.150 below Points Per Play and more than 0.300 below EqPts+. That was unexpected.
In the past few weeks, you have seen the rollout of a few new concepts that were key to establishing this Varsity Numbers column. You have also seen the potential statistical importance of numbers like EqPts, Points Per Play (PPP) and S&P (Success Rates + Points Per Play). Future columns will attempt to apply some of these concepts to what you are seeing unfold in the 2008 college football season. Keep the good feedback coming, it has been quite useful.
3 comments, Last at 11 Oct 2008, 2:26pm by War Eagle