Handicapping 2010 NCAA Week 11

We can use TopoRank to reverse-engineer the expected point differential between two teams. Let's see how it does for this week's Top 25 matchups. The predicted winner is bolded.

Visitor	Home	Prediction	Spread	Actual
1. Oregon	California	12.2	19-20.5	2
Georgia	2. Auburn	14.2	7.5-8.5	18
San Diego St	3. TCU	25.9	27-28	5
4. Boise St	Idaho	20.9	33.5-34.5	38
LA-Monroe	5. LSU	32.3	31-34	skip
6. Stanford	Arizona St	12.9	4.5-6.5	4
Indiana	7. Wisconsin	16.4	21.5-22.5	63
Kansas	8. Nebraska	24.7	34-34.5	17
Penn St	9. Ohio St	13.9	17.5-20.5	24
10. Oklahoma St	Texas	8.0	4-6.5	17
19. Mississippi St	12. Alabama	12.2	13-14	27
13. Iowa	Northwestern	11.5	9.5-12	skip
14. Utah	Notre Dame	6.4	5-6	-25
UTEP	15. Arkansas	30.1	28.5-30	31
Texas Tech	16. Oklahoma	13.9	16-16.5	38
24. Kansas St	17. Missouri	7.2	13-13.5	10
USC	18. Arizona	5.1	4-5.5	skip
20. Virgina Tech	North Carolina	3.7	3-5.5	skip
21. Nevada	Fresno St	7.4	8-9.5	1
23. South Carolina	22. Florida	3.7	6-7	-22
25. Texas A&M	Baylor	5.5	3-3.5	12

I'm also tracking approximate spreads for these games and seeing how TopoRank would do. In each case, if TopoRank predicts a bigger win than the spread, I'd bet for the spread (and vice versa). This is slightly complicated by the fact that spreads vary by institution, so if TopoRank's prediction falls within a range of published spreads, the game will be ommitted ('skip').

This week, TopoRank made the right call 10 out of 17 times for a success rate of 58.8%. Assuming someone, somewhere, would pay 21:11 for a correct pick, the rate of return would be 12.3% ...

2010 College Football Rankings, Week 11

TopoRank Top 25 after Week 11:

1. Oregon

2. TCU

3. Stanford

4. Boise St

5. Auburn

6. Ohio State

7. Alabama

8. Arkansas

9. Missouri

10. Oklahoma St

11. Oklahoma

12. Nebraska

13. Texas A&M

14. Iowa

15. Arizona

16. Virginia Tech

17. LSU

18. Florida

19. Utah

20. Kansas St

21. South Carolina

22. Michigan St

23. Wisconsin

24. Southern Cal

25. Nevada

The BCS results aren't out yet, but I expect to see some big differences again. Oregon and TCU stay on top. Compared to last week, we see Missouri, Utah and Arizona take big falls after ugly losses, and Stanford and Boise State jump to #3 and #4 with big wins. TopoRank still shows Auburn trailing these teams to round out the top 5.

If the season ended today, Oregon would face Auburn in the BCS Championship game, even though TCU and Boise State are more deserving. This is the strength of perception, especially about the SEC.

2010 College Football Rankings, Week 9

I am using TopoRank to rank college football teams, and have compared my Week 9 standings and the BCS standings. I'm calling out the teams that ended up at least 5 spots lower or higher and examining if these re-rankings make sense. (*I've also added in the combined of two TopoRank algorithms)

Rank	BCS	TopoRank	*TopoRank Combined
1	Oregon	TCU	Oregon
2	Auburn	Oregon	Missouri
3	TCU	Oklahoma	TCU
4	Boise State	Alabama	Oklahoma
5	Utah	Missouri	Nebraska
6	Alabama	Stanford	Alabama
7	Nebraska	Boise State	Stanford
8	Oklahoma	Utah	Auburn
9	Wisconsin	Iowa	Boise State
10	LSU	Arizona	Iowa
11	Ohio State	Auburn	Arizona
12	Missouri	Nebraska	Ohio State
13	Stanford	Ohio State	Utah
14	Michigan St	Arkansas	Arkansas
15	Arizona	Virginia Tech	Virginia Tech
16	Iowa	Michigan St	Oregon St
17	Oklahoma St	Oregon St	South Carolina
18	Arkansas	Illinois	Oklahoma St
19	South Carolina	South Carolina	LSU
20	Mississippi St	Oklahoma St	Michigan St
21	Baylor	Mississippi St	Florida St
22	Virginia Tech	Florida	USC
23	Nevada	Florida St	Mississippi St
24	Florida St	USC	Florida
25	NC St	LSU	Texas A&M

The Good (*-no longer 5 spots better):

Oklahoma* (BCS #8, TopoRank #3) - This is a stretch. Oklahoma's only good win was a demolition of Florida St, and narrow wins over Utah St, Air Force, Cincinnati don't paint a picture of a national powerhouse. An alternate version of TopoRank has them at #8 and I think that's much more correct. Perhaps I also need to average the results of different algorithms (the subjectivity in TopoRank is how much credit to give a team over another based on the final score differential).

Missouri (BCS #12, TopoRank #5) - Missouri was racking up dominating wins and a close win over a probably underappreciated (now 6-2) SDSU team when it ran into a backup running back who had a career game. This is more a case of a very strong team tripping a bit than a consistently semi-strong performer. I think the #5 ranking is much more justified than #12.

Stanford (BCS #13, TopoRank #6) - This team packs a major punch and has absolutely blown up 6 of its 8 opponents (Washington St made the score look a lot closer in the closing minutes). The Cardinal get it done and don't leave things to chance. Their one loss is against one of the best teams in the last decade, and they managed to respond late in the game to pull off a win against borderline-Top-25 USC. Name 6 teams you'd favor over them. I sure can't.

Iowa (BCS #16, TopoRank #9) - Iowa is bearing the brunt of having 2 losses, both of them being close games against good teams. Their wins have been mostly beatdowns, including a 37-6 trampling of then-#6 Michigan St. This team is strong and would be competitive against many of the teams ranked ahead of them.

Arizona* (BCS #15, TopoRank #10) - Arizona is smothering opposing offenses, scores points in bunches, and is one subpar half away from being 8-0. Their defense is consistently stepping up when the offense stalls. This team is never out of a game.

Virginia Tech (BCS #22, TopoRank #15) - A close loss to a top team followed by a letdown loss on 5 days rest has doomed this team. They've rallied to win 6 straight since, their best win being a comeback against #23 NC St. Perhaps the Hokies are benefiting too much from score differential. The other algorithm (which gives less credit for blowouts) has them at #18.

Oregon St (BCS unranked, TopoRank #17) - The Beavers took on two national title contenders and were overmatched in both games, but not badly. And Jake Locker had one of his 2 good games to pull out a double overtime win. Their big win is against Arizona, but really they're a victim of their schedule. Replace games against TCU and Boise St with weaker opponents and this team is likely 6-1 and in the thick of the 10-15 range.

Illinois* (BCS unranked, TopoRank #18) - 3 losses against very good teams (including then unranked Missouri) have made voters forget about this team. However, they turned in dominating performances in 4 of 5 wins, and 2 of the 3 losses to good teams should be considered competitive games. This is not the profile of an unranked team. Perhaps they should be borderling Top 25. The other algorithm currently places them #32, so an average of the two seems to place them just right.

The Bad (* - no longer 5 spots worse):

Auburn (BCS #2, TopoRank #11) - Yes, Auburn is 9-0. Yes, they have big wins over South Carolina, Arkansas and LSU. But, Auburn barely squeezed by SC and LSU, and pulled away from Arkansas after Ryan Mallett was hurt. And they had narrow wins over barely ranked Mississippi St and unranked Clemson and Kentucky. This paints a picture of a team that is certainly good, but sweating out wins against teams across the Top 25 and beyond.

Nebraska* (BCS #7, TopoRank #12) - Nebraska has 2 wins against good teams, a win that used to look good (Washington), 4 wins that carry no weight, a loss that looks increasingly bad (Texas). This team created a strong impression against weak competition, then pulled out 2 wins powered by big plays against good teams. This team has gotten a little bit lucky, and is not as consistently good as voters think.

LSU (BCS #10, TopoRank #25) - This team might actually be ranked closer to its true strength if Tennessee hadn't tried to defend with 13 men on the field. In addition to the biggest gift of the year and a solid win over Mississippi State, they've pulled off narrow wins over no-longer-ranked teams North Carolina, West Virginia and Florida, and never-ranked Vanderbilt and McNeese State. Add in a loss to overranked Auburn, and this team belongs in the tail end of the Top 25.

The Additional Bad by Combined TopoRank:

Boise State (BCS #4, Combined TopoRank #9) - With only 2 quality opponents, it's hard to argue that Boise State has shown it's a national contender. We specifically see the Combined TopoRank place less emphasis on the margins of victory, causing the Broncos to fall lower.

Utah (BCS #5, Combined TopoRank #13) - The Utes suffer for the same reason as Boise State in this assessment, and their blowout loss to TCU hurts them further. With their only other quality game being against no-longer-ranked Pitt, it's hard to make the case that Utah should be a top 5 team.

Conclusion:

I think TopoRank places several teams more correctly with respect to their actual strength. I'll modify the rankings to incorporate both algorithms and republish* the results with Week 10 included. While it certainly is true that a team needs to prove itself by winning games (and not just playing good teams close), I contend teams are overly penalized for losses. Even if TopoRank can't overcome conventional thinking and 'gut feel', perhaps it can make a good handicapping tool for those of us headed to Vegas....

* I have updated the table with the Combined TopoRank results. I think they highlight the same over and underrated teams, and correct some of the extreme jumps. It also highlights the uncertainty of Boise State and Utah's rankings.

The Long Field Goal

Your team has just stalled their drive at the 35 yard line, and your kicker has a pretty good leg. Send on the field goal unit, right? We're seeing teams sometimes elect to punt or go for it... which is right?

Using average points per drive numbers, we see that pinning the opponent deep means they can expect to score about 0.7 points on the ensuing drive. Missing the field goal pushes this number up around 3.0 for a net loss of 2.3. Even if you make the field goal, your opponent will average 1.6 points on the next drive, netting 1.4 points. Kickers convert on long field goals about 50 percent of the time, meaning the likely outcome is a loss of 0.8 points per attempt. In other words, the long attempt is about as good as pinning the opponent deep, especially when factoring in that your punter might put the ball in the endzone.

But, what about going for it? Getting the first down ups the average score to 3.6, for a net of 2.0 after factoring in the ensuing drive from the other team. Failing to convert yields the same 3.0 to the other team as missing the field goal does, so the conversion would have to succeed about 60 percent of the time to break even.

Turns out that all the options are pretty even, and this is really an area where a coach can make their money by having a good feel for the right play to call.

Buying 4 points

Teams try all manner of improving the defense, but it's a complex machine with many moving parts. What if there were an easier way to keep points off the board? A ball-control offense does the trick, but that tends to score fewer points.

Turns out there's a very simple, side-effect free, but hypothetical way to do this. The Cowboys got it half right last year. I've charted every drive from the first 20 games of this year and quantified a fairly obvious truth: the farther back a team starts its drive, the less likely it is to score. When a team starts inside its own 10, it averages about 0.67 points per drive. When it starts past its own 40, that number rockets up to 3.0. Here are the ranges and average points scored:

1-10: 0.67

11-20: 0.70

20 yard line: 1.00

21-30: 1.60

31-40: 1.75

40+: 3.0

The average starting field position for a team accepting a kickoff is its own 26. 2 yard line.

Suppose we could find a player whose only job it was to kick a ball through the endzone, time after time. The average team in 2009 kicked off just about 5 times a game. According to the chart above, we'd shed about 3 points allowed per game!

Charting offensive yards vs points scored yields a highly correlated best-fit line showing that a team scores 1 point for about every 9.1 yards over 143 they gain in a game. So, a team gaining 250 points can expect to score about 12 points in a game, 360 yards yields 24 points, and 500 yards comes out to around 40. By this metric, a savings of 6.2 yards, 5 times a game, saves about 4.5 points. Even better! (but we'll use the first method)

Want a bonus? Since our kickoff duties have been offloaded to Mr Cannonleg, our field goal kicker can be optimized for accuracy. Teams attempted an average of 1.8 field goals a game, and converted at an 81 percent rate. The best kickers in the league hit about 10 percent more often than that, resulting in an extra 0.54 points per game. Replacing an inaccurate kicker? The bottom of the league is another 10 percent below the average, so the net effect would reach 1 point per game.

And sprinkles on top (though now we're being a bit silly)? The concept extends to punts too. Teams punt 4.8 times per game, of these 1.6 are downed inside the 20 and 0.4 result in touchbacks. If your punter can keep those touchbacks out of the endzone, that's another 0.12 points. Drives starting after punts yield 1.36 points. Mat McBriar was able to punt past the 20 57% of the time, which was 15% better than the league average. If the average team could improve by that much, they could expect to give up 0.72 points less per game. In short, improved punting could improve your team's defense by over 0.8 points per game. Can your cannon-legged kicker always get a touchback? The benefit jumps to 1.73 points per game! Even if your average punter is replaced by McBriar, about half a point is saved.

Points saved per game over average for

always kicking a touchback: 3.0

having an accurate field goal kicker: 0.5

* always punting a touchback: 1.7

* always punting inside the 20: 2.7

being best at punting inside the 20: 0.5

*- these are kinda ridiculous ...

Now for the hypothetical part: there has never been a kicker who could kick that far, that often. I suspect there are some soccer players out there who could achieve results close to these, especially in warm weather/indoor stadiums. They could probably be signed for minimum contracts and the benefit is probably bigger than keeping an extra special teams or backup player on the active roster. How much bigger? There's a very strong correlation: for every 2 point differential a team averages, it gains a win. 3 points, 1.5 wins per season. Tell me that's not worth it.

Draft Metrics: Linebackers

Linebackers are the multi-tool of the defense. Some come blitzing off the edge, others fill running lanes, while some frequently drop back in coverage. The wide variety (or combination) of objectives makes them hard to compare, so I'll be leaning on pro-football-reference.com's Career Approximate Value (CarAV) to help me rank players. Similar to defensive backs, I've broken down picks between 2000 and 2008 into starter quality (starts at least 80% of seasons), journeyman (at least 50% of seasons) and the rest. I've correlated these with CarAV ranges to get an idea of what level of play is required for each bucket, and will look at player yields from the draft using these points of reference.

By inspecting the list of players, the 'elite' range for CarAV is above 9, and 'good' is above 7.5. Starter quality players average about 6, so I'll consider this the cutoff for an 'adequate' starter. The bottom third of the journeyman range is 4.5, which will become the cutoff for a bust ('backup'). Now let's see what the data shows.

Round 7:
Of 45 picks, there are 2 backup level players.

Round 6:
Of 37 picks, there are 1 good, 2 adequate and 1 backup players.

Round 5:
Of 44 picks, there are 3 adequate and 3 backup players.

Round 4:
Of 42 picks, there are 5 adequate and 4 backup players.

Round 3:
Of 53 picks, there are 1 elite, 1 good, 4 adequate and 6 backup players.

Round 2:
Of 39 picks, there are 3 elite, 2 good, 8 adequate and 5 backup players.

Round 1:
Of 27 picks, there are 5 elite, 6 good, 6 adequate and 5 backup players. 5 players were busts.

Conclusion:
We see the similar trend of little differentiation in rounds 5 through 7. Less than 5% of picks are of adequate starting caliber, and only another 5% are destined to backup duty. In rounds 3 and 4, the odds of each double, but even building depth is a slim proposition. Things change drastically from here though, as nearly half of 2nd round picks seeing significant action, with 1 in 3 being starter quality. We also see about 1 in 5 picks end up in the good or elite category. Round 1 yields about 2 starters per 3 picks and those are equally likely to be good or elite. The bust rate is less than 1 in 5.

Based on past pick history, teams should not use a top-10 pick on a linebacker. Brian Urlacher is the only good or elite pick in that range (#9). The other 4 (Arrington, Hawk, Sims, Rivers) have been quality players, but didn't reach the higher status for one reason or another (Arrington and Rivers missed significant time to injury). Looking more closely at the draft history, the quality starts to drop off pretty significantly by the middle of the 2nd. The range is just right for every team to get one good shot at a starting linebacker.

Draft Metrics: Defensive Backs

What good is a strong pass rush if there aren't players to make a quarterback pay for poor passes? Defensive backs are the last line of defense, and have the largest ability to make big plays on the ball, but they're also the ones who get singled out for being burned over the top. So, what can we expect from a player? Most seasons, the NFL leader has 8-10 interceptions. The names are familiar: Darren Sharper, Ed Reed, Antonio Cromartie, Champ Bailey ... and yet, that level of success is incredibly hard to maintain. In fact, the highest per-season interception average over the last decade is just 5.8 (Ed Reed), and only two other players (Asante Samuel and Dominique Rodgers-Cromartie) are at 5 per season. 7 more players are between 4 and 5, and there aren't quite enough players total (28) to leave every NFL team with one player averaging at least 3 interceptions per year.

We'll define players averaging 4 or more picks as 'elite' playmakers, 3 or more as 'good', 2 or more as 'adequate' and the rest are freely replacable. Any player starting over 80% of his seasons (to allow for one injured or bad season) in the league will be labeled a 'career starter', and players starting at least 50% of the season will be labeled 'fringe starters'. Of 485 picks made between 1999 and 2008, only 48 meet this criteria. So, based on picks made from 1999 to 2008, where should teams look to find these players?

See full data here.

Round 7:
Of 70 picks, 3 fall into the adequate playmaker category and all others are replacable. 1 player (Finnegan) has made 1 Pro Bowl. 1 player is a career starter and 6 are fringe.

Round 6:
Of 71 picks, 2 are adequate playmakers and all others are replacable. 1 player (Bethea) has made 1 Pro Bowl. 2 players are career starters and 1 is fringe.

Round 5:
Of 69 picks, 4 are adequate playmakers, and 1 player (Azumah) has made 1 Pro Bowl. 3 players are career starters. 3 are career starters, 4 are fringe.

Round 4:
Of 79 picks, 1 (Samuel) is an elite playmaker, 3 more (A.Henry, Vasher, Rhodes) are good and 5 more are adequate. There are 5 Pro Bowl appearances from this group (Samuel: 3, Vasher: 1, McGee: 1). 5 are career starters, 9 are fringe.

Round 3:
Of 67 picks, 2 (Atogwe, Wesley) are good playmakers, and 6 more are adequate. Wilson has made 3 Pro Bowls, C.Hope has made 1. 7 are career starters, 8 are fringe.

Round 2:
Of 73 picks, 1 (Mathis) is elite, 5 are good, and 14 are adequate playmakers. There are 13 Pro Bowl selections from this group, though the 2 belonging to Devin Hester are not for playing defense. 10 are career starters, 32 are fringe.

Round 1:
Of 56 selections, 8 are elite, 7 are good and 14 are adequate. 47 Pro Bowl spots over the last decade are from this group. 20 are career starters, 19 are fringe.

Defensive backs show a fast dropoff in playmaking skill and starting caliber through the rounds. Rounds 5-7 are essentially interchangable, yielding 6 career starters, 3 Pro Bowls and no good or elite playmakers in 210 tries. Rounds 3 and 4 contain 12 career starters, 19 fringe, 8 Pro Bowls, 1 elite and 5 good playmakers in 146 tries. The odds of an impact player stay safely under 10 percent. Round 2 improves the odds, but there's still only about a 30 percent chance of picking up a playmaker, and only about half of those stick as career starters.

Round 1 drastically improves the odds of picking up a playmaker who becomes a starter over a career. Bust rates are very high, barely 1 in 3 first rounders reach this level. Another 1 in 3 are shuffled in and out of lineups, relegated to Mel Kiper's assessment that they are not the long-term future of the position and an upgrade could be made. But Mel, the odds of the new pick beating them out just isn't that high.

It seems drafting defensive backs is a constant struggle, and teams picking well need to treasure that player. Even for just quality depth, picks should come from the top 2 rounds. Because of the high risk involved, these players should not be picked so high that their contracts become prohibitive. A smart team must realize there's a good chance their selection will not pan out, and has to be ready to move on in a few years.