College Football 2011

Can a defense force turnovers? A look back at Stanford, Oklahoma State.

Heading into the Fiesta Bowl against Stanford, Oklahoma State led the nation with 42 turnovers created by their defense. But is this a skill? Can a defense consistently force turnovers? We searched for any statistical analysis that could answer this question. The most complete study was done by Bill Barnwell at Grantland. He found that turnovers in the first 5 games of the NFL season is weakly correlated with turnovers in the remaining 11 games (correlation coefficient of 0.14). We used this study in our Fiesta Bowl preview to suggest that a large turnover total is not predictive of future turnovers.

So how did the Fiesta Bowl turn out? In the 1st quarter, Oklahoma State cornerback Justin Gilbert intercepted an Andrew Luck pass. It’s reasonable to argue that the Cowboys forced this turnover, as Gilbert had the speed and agility to step in front of the Stanford receiver. In the 3rd quarter, Oklahoma State recovered a Stanford fumble. The replay clearly showed that Andrew Luck had a bad exchange with Geoff Meinken. In no way did Oklahoma State force this turnover. The defense only capitalized on a mistake by the offense. Stanford had 2 turnovers the entire game, nothing near the almost 4 turnovers per game the Oklahoma State defense had received this season. It’s not that Oklahoma State didn’t try. On two catches in the 2nd half, Stanford receiver Griff Whalen had a horde of defenders standing him up and trying to rip the ball away. No luck.

At the end of regulation, Stanford missed a field goal to win the game, and Oklahoma State won in overtime. The outcome supports The Power Rank’s assertion that Oklahoma State was the better team.

The final college football rankings for the 2011 season.
But luck was also involved, as Oklahoma State converted a critical field goal after the Stanford fumble in the 3rd quarter. Looking ahead to next year, analytics suggests that randomness plays a huge role in how many turnovers a defense forces. No matter how much they practice creating turnovers, don’t be surprised if Oklahoma State doesn’t force 44 turnovers next year. It might be a more difficult season for the Cowboys.

For more content, find The Power Rank on Twitter.

—Can a defense force turnovers? A Stanford, Oklahoma State preview.
—About The Power Rank.
—College football’s incredibly slow progress towards a playoff.
—The Power Rank featured on KALX Spectrum, the science and technology show on UC Berkeley student radio.

3 last thoughts on LSU and Alabama

By Dr. Ed Feng Leave a Comment

After a weekend of the BBVA Compass Bowl and the GoDaddy.com Bowl, the national championship game is finally here. On Friday, we looked at what analytics said about Alabama Nick Saban’s 4th down decision making in the first game between Alabama and LSU. Here are 3 more thoughts about the rematch.

1. Alabama and LSU have two historically good defenses.

Alabama and LSU have distinguished themselves on defense among the 120 bowl subdivision teams.

The visual shows our rankings that account for strength of schedule in scoring offense and defense. Alabama and LSU have clearly distinguished themselves among the 120 bowl subdivision teams this year, posting a 9.36 and 9.61 points per game in adjusted scoring defense. We got interested in whether any defenses over the last few years have come close to these numbers. No. Since 1997, only the 2000 Florida State defense (9.77) has come within a point of LSU and Alabama.

2. LSU has led the nation in turnover margin.

According to cfbstats.com, LSU has forced 30 turnovers while coughing up the ball only 8 times. This +22 turnover margin leads the nation, even ahead of Oklahoma State. However, we have previously written about how early season turnovers are weakly correlated with late season turnovers. So a large turnover margin is almost meaningless in predicting performance against Alabama tonight. In the future, we will attempt to account for this in our rankings to make them more predictive. At this point, we can only say that LSU might be overrated in The Power Rank because of this huge turnover margin.

3. If team A has a better defense and quarterback…

If team A has a better defense and quarterback than team B, then team A will win on average. Seems simple enough. Well, team A is Alabama. Alabama’s defense has topped our rankings all year. LSU only closed the gap after Georgia Southern gave Alabama fits with their triple option attack. At quarterback, LSU’s Jordan Jefferson has completed 59% of his passes over his career while Alabama’s AJ McCarron has posted a 67% completion rate this year. Moreover, Alabama and running back Trent Richardson have the edge in the running game, as they have run for 5.58 yards per carry compared to 4.96 for LSU.

And the winner is…

The last two thoughts suggest that a 60% chance for an LSU victory might be too high. In fact, the line has shifted from LSU by 1 to Alabama by 2. The Power Rank puts the most likely outcome around LSU by 4.2 points. Have an opinion on this tight game? Please leave us a comment.

Thanks for reading.

For more content, find The Power Rank on Twitter.

Should Nick Saban have gone for it on 4th down? A LSU, Alabama (p)review.

By Dr. Ed Feng Leave a Comment

Alabama marched into LSU territory plenty of times. In their early November showdown, Alabama got to the LSU 27 and 33 yard line on their first two drives, only to have Cade Foster miss two field goals. The next Alabama drive stalled at the LSU 31, and this time coach Nick Saban gave Jeremy Shelley, usually their short field goal kicker, a shot from 49 yards. LSU blocked this attempt. In their epic defensive showdown, Alabama outgained LSU in total yardage (295 to 239) but ended up losing 9-6 in overtime. With all those deep penetrations into LSU territory, should Saban have gone for it on 4th down?

Here, we list the 4th down situations that confronted Saban, along with the result.

4th and 7 at LSU 27; Foster missed a 44 yard field goal attempt
4th and 20 at LSU 33; Foster missed 50 yard field goal attempt
4th and 17 at LSU 31; Shelley had 49 yard field goal attempt blocked
4th and 8 at LSU 17; Shelley made 34 yard field goal
4th and 4 at LSU 29; Foster made 46 yard field goal

So what does analytics say about 4th down decision making? Brian Burke, founder of Advanced NFL Stats, has done the most complete analysis of the subject based on the NFL. With no similar analysis of college football, we’ll look at Saban’s decisions through this work. Clearly, this is far from ideal, and we’ll address these concerns later.

Expected Points

To understand Burke’s analysis, we must first introduce the idea of expected points. For a situation, such as 1st and 10 from the offense’s 27 yard line, the expected points is the average net points for the offense on the next score. If a team gets a 1st and 10 from the 27 a million times, sometime they will score a touchdown for +7 points. Other times, the defense will hold, resulting in a field goal for the other team, or -3 net points for the offense. Perhaps, the defense even returns an interception for a touchdown (-7 points). But an average over all these scores gives a net 0.7 points for the offense. Burke arrived at 0.7 by looking at every 1st and 10 starting at the 27 from the 2000-2008 NFL seasons and averaging the next score in the game. For the expected points for a 1st down at every yard line, see this figure.

These values for expected points allow Burke to evaluate 4th down decision making. This is easiest to understand with an example, such as the 4th and 4 from the LSU 29 that Alabama faced. Alabama has a 53% chance of making a 1st down, which we’ll assume will spot the ball at the LSU 25. This is worth 3.6 expected points. In the 47% chance that Alabama doesn’t make the 1st down, we’ll assume LSU gets the ball at the 29 yard line. This is worth -0.9 expected points for Alabama. To calculate the expected points, we add the probability of each choice times the expected points for each outcome. In this case, (0.53 * 3.6) – (0.47 * 0.9) gives 1.46 expected points for Alabama.

For a field goal attempt, the average NFL kicker will convert 67% of field goal attempts from 46 yards out. This is actually worth 2.3 expected points, since we must subtract 0.7 from 3 because the kickoff gives the opponent the ball at their 27 yard line. Missing the field goal gives LSU the ball at the 29 yard line, again worth -0.9 expected points for Alabama. Performing the same analysis as before gives 1.24 expected points for kicking the field goal. Since 1.24 is less than 1.46, Alabama should have gone for it on 4th down. This analysis is summarized in this plot of line of scrimmage versus yards to go on 4th down.

Analytics says that Saban was…

Based on Brian Burke’s work on the NFL, we just showed that Nick Saban should have gone for it on 4th and 4 from the LSU 29. However, this analysis implies that Saban should NOT have gone for the 1st down on the other four 4th down situations. He made the right call to kick a field goal. In each of these four situations, his offense faced 7 or more yards to the first down. This distance presented a daunting challenge, especially against LSU’s defense.

Of course, Saban shouldn’t hitch his coaching career to this study. Many factors make it difficult to apply analytics to actual 4th down decisions. First, Burke looked at NFL games, and many quantities such as 4th down conversion rates could differ in college football. Second, Burke’s analysis averages over all teams, which doesn’t account for how a good Alabama offense (12th in our offense rankings) will match up with an exceptional LSU defense (2nd in our defense rankings behind Alabama). Lastly, the quality of the kicker has a huge impact on the analysis. From the outside, Cade Foster only attempts field goals from 40 yards or greater. While he’s only 2 of 9 this year, he banged home 7 of 9 last year. On the inside, Nick Saban probably knows best the likelihood that Foster will make a 50 yard field goal. Analytics doesn’t have all the answers and never will in a complex team sport like football.

National Championship Game Preview

However, analytics can provide useful guidelines for 4th down decision making. If you look at Burke’s 4th down decision plot, a team should almost always go for it with 2 or less yards to go. This is true as far back as the 28 yard line, the spot in which New England coach Bill Belichick famously failed to convert a 4th and 2 against Indianapolis. For Nick Saban on Monday night, he might consider going for it on 4th and 2 at the LSU 20. The Tiger’s defensive line is stout, but Alabama features Outland Trophy winner Barrett Jones on their offensive line and Heisman finalist Trent Richardson at running back. Moreover, quarterback AJ McCarron has completed almost 67% of his passes this year. Alabama has all the tools to score more than 6 points against LSU’s defense.

The Power Rank predicts LSU has a 60% chance of winning. What do the other 40% of outcomes look like? LSU excels at every position on the field except quarterback. Despite his athleticism, Jordan Jefferson has completed 59% of his passes over his career. His backup Jarrett Lee has posted a 56% completion rate over his career, although he has completed 62% of his passes this year. Lee came into the Alabama game earlier this year with an absurdly low interception rate. We forecasted that this low rate would not continue, and he threw 2 picks against Alabama. A similar performance by Jefferson on Monday night could tip the game for Alabama.

For more content, find The Power Rank on Twitter.

Can a defense force turnovers? A Oklahoma State, Stanford preview.

By Dr. Ed Feng Leave a Comment

Oklahoma State forced 5 turnovers in 44-10 rout over Oklahoma in this year’s Bedlam rivalry game. The turnovers had such an impact that Oklahoma didn’t score a touchdown until late in the 4th quarter despite racking up 358 yards. Oklahoma State preaches creating turnovers in practice, as players are required to repeat plays or run afterwards if they do not force enough turnovers. Defensive coordinator Bill Young’s units have forced 42, 34 and 30 turnovers in his 3 year tenure, well above the 22 turnovers that an average bowl subdivision defense gets a year.

But can a team really force turnovers?

An analytic approach to turnovers

Bill Barnwell at Grantland has done the most complete study of forcing turnovers. After the Pittsburgh Steelers only forced 2 turnovers in their first 5 games this year, Barnwell asked whether turnovers in the first 5 games can predict turnovers in the remaining 11 games. To answer this question, he looked at all NFL teams from 1990 through 2010 and calculated the correlation coefficient between early and late season turnovers. This coefficient is related to a scatter plot in which each point represents a team’s turnovers in the first 5 games and last 11 games in its horizontal and vertical coordinate respectively. The correlation coefficient, which ranges from -1 to 1, describes the scatter of this plot. See Wikipedia for pictures corresponding to values of the coefficient. A correlation coefficient of 0 implies high scatter with no trend between the two variables.

For his NFL study, which included over 600 data points, Barnwell found a correlation coefficient of 0.14, a very weak correlation between turnovers in the first 5 games and last 11 games. To think about this a different way, the square of the correlation coefficient is a measure of how much of the variation in the turnovers the last 11 games is explained by the turnovers in the first 5 games. Only 2% of this variation is predicted by early season results. For example, after getting only 2 turnovers in their first 5 games, Pittsburgh has forced 12 turnovers in their last 10 games.

Barnwell’s study suggests that Oklahoma State’s high turnover total will not correlate with their Fiesta Bowl performance. Turnovers are not that useful in predicting future performance. While Oklahoma State’s defense is 9th in our defense rankings, which is scoring defense that accounts for strength of schedule, this might be too high because of their good fortune in turnovers.

Offense and Defense rankings for Stanford and Oklahoma State in the Fiesta Bowl

How Oklahoma State’s defense will match up with Stanford

To better understand how Oklahoma State’s defense will match up with Stanford’s 5th ranked offense, let’s break down rushing and passing separately. According to cfbstats.com, Oklahoma State’s defense gives up 4.34 rush yards per carry, 72nd worst in the country. This softness was apparent on a cold night in Ames. In overtime, Oklahoma State needed to prevent Iowa State from scoring to force another overtime. Instead, they gave up a touchdown on 3 straight running plays. Of course, this poor defense might have been caused by an urgency to force a fumble. But this just shows that a defense can’t necessarily force a turnover when it needs it most. Stanford has a run first offense that gains 5.33 yards per play. Quarterback Andrew Luck barely threw a pass when the running game racked up 10.1 yards per carry against Washington. If Stanford can get an early lead, Oklahoma State’s defense will see a steady diet of running plays.

Through the air, Oklahoma State’s gives up 6.2 yards per pass attempt, 15th best in the nation according to cfbstats.com. They give up fewer yards per attempt than Oregon, a defensive unit that gave Stanford fits in the passing game. After that game, Bay area media would not stop talking about how Cardinal receivers couldn’t get separation from Oregon’s secondary. Life might be difficult for Andrew Luck and the Cardinal offense if they fall behind and have to throw on the majority of plays.

The most recent team rankings predict Oklahoma State by 6.6 over Stanford. We’ll see how turnovers affect the actual outcome of the game.

For more content, find The Power Rank on Twitter.

—About The Power Rank.
—Ranking college football bowl games: a viewer’s guide.
—College football’s incredibly slow progress towards a playoff.
—3 types of college football coaches that got fired.
—The Power Rank featured on KALX Spectrum, the science and technology show on UC Berkeley student radio.

Can Cal beat Texas? The most important matchup of the Holiday Bowl

By Dr. Ed Feng Leave a Comment

Recently, we started ranking games according to the strength of the teams and an excitement factor based on our predictions. When applied to this year’s Bowl games, the Holiday Bowl matchup between Texas and California came up 6th out of 35, just behind the Cotton Bowl game between Kansas State and Arkansas. While it might seem strange that a December 28th game between two 7-5 teams ranks so highly, the matchup has a high excitement factor. The Power Rank (TPR) predicts a 0.8 point victory for Texas in a close game, essentially a coin flip. In fact, the algorithm predicted a 0.3 point win for Cal prior to the start of Bowl season. It changed its mind after Arizona State’s 56-24 loss to Boise State last week, making Cal’s 9 point win over the Sun Devils less impressive.

Also, after a week of watching teams like Florida International and Marshall play in bowl games, the Holiday Bowl does not disappoint in the team strength department. While we mocked the AP poll for putting Texas 19th after their 3-0 start, the Longhorns made it as high as 21st in TPR. Losing 3 of their last 4 dropped Texas to 26th at the end of the regular season. Cal was 62nd after an embarrassing 17 point loss to UCLA . They ended the season 3-1, and their only loss was a 3 point game at Stanford in Big Game. This late season surge pushed Cal up to 28th at the end of the regular season, a big factor in making the Pac-12 our third rated conference behind the Big 12 and SEC.

To further break down this matchup, we look at our offense and defense rankings.

This visual is based on offense and defense ratings that correspond to the predicted points scored or allowed against an average opponent. (Click on offense and defense for the numerical values.) Texas has a better defense while Cal has the better offense. These rankings predict a 24.0-22.7 win for Texas, a defensive battle considering that a team scores 26.9 points per game on average. The 1.3 point margin of victory is similar to the 0.8 predicted by the team rankings.

These rankings point towards the Cal offense against the Texas defense as the most important matchup of the game. After watching his quarterback Zach Maynard struggle early in the season, Cal coach Jeff Tedford started calling more running plays in the last 4 games. The Bears racked up over 5.9 yards per carry in all but one of those games. This proficiency in the running game helped Maynard, who completed 67% of his passes and threw 5 touchdowns to 1 interception in those last 4 games. However, running the ball against Texas will not be easy, as the Longhorns only allowed 3.34 yards per carry, good for 18th in the country according to cfbstats.com. If Maynard can avoid the interceptions that plagued Cal early in the season, the Bears will stay in the game. If Maynard can find receivers Keenan Allen and Marvin Jones down the field, Cal can beat Texas.

For more content, find The Power Rank on Twitter.

Get a sample of my best football predictions