• Facebook StumbleUpon
• Email

1. Scipio Tex said:

   September 24th, 2008 at 8:03 am

   Agreed. I would also add turnovers to that list and then I’m good.

2. Big Satan said:

   September 24th, 2008 at 8:16 am

   Interested in seeing how this defense responds against much stiffer competition before we say this defense is a lot better than it was a year ago. How about a year ago through our non-conference schedule?

3. Kafka said:

   September 24th, 2008 at 8:17 am

   Are the comparisons between this year and last year include all of last year? I am guessing that UT’s D stats looked much better after 3 games last year than they did by the end of the 2007 season (because the non conference schedule at the beginning of the season is easier than the conference games that come later).

   A stat I would like to know is the average time a QB can hold onto the ball before he is forced to throw the ball. Another stat is how long can the QB hold the ball before he is forced to move. I’d like to know how often the D creates a disruptive play that puts the O behind schedule re: down vs distance. It would be interesting to know how often the D calls the right play/formation versus the O.

4. HenryJames said:

   September 24th, 2008 at 8:34 am

   Through the first 3 games of last year:

   YPA: 6.11
   3rd Down: 35%
   Red Zone: 57%
   Scoring defense: 19.3

   So we’re not a lot better, but we are better.

5. kriess said:

   September 24th, 2008 at 8:36 am

   Yeah those stats are impossible to gather due tot he fact that every QB has diff tendencies. We all know it may take Colt a few weeks to let go of the football. Secondly how are we going to know if its the “right” defensive formation called against an offense? If you know the correct play to call against every O formation then you need to be on the sidelines with a headset. If maybe Muschamp can get this stat together it would hold some weight, but any joe schmo or “statistician” really is out of his element on this one.

6. Mister Mike said:

   September 24th, 2008 at 8:38 am

   “So we’re not a lit better, but we are better.”

   You guys became a helluva lot better when you put the big red rubber NO FUCKING WAY stamp on Akina as DC and went out and got Muschamp instead.

7. Vasherized said:

   September 24th, 2008 at 9:01 am

   Once our DBs learn to turn around and make a play on the ball — assuming they are within ten feet of the WR to begin with — I’ll be more optimistic about our turnover ratio.

   Beasley’s regression is perplexing. Hopefully it’s injury related and something he gets past quickly. Palmer certainly isn’t the answer.

8. dedfischer said:

   September 24th, 2008 at 9:12 am

   3rd down defense has been the biggest improvement for Tech and I think we’re 3rd in the country at 22%. However, I think it’s because we’ve been stopping the run on first down better, which is the key to playing good 3rd down defense.

9. kriess said:

   September 24th, 2008 at 9:27 am

   And bc TT was playing UMASS school for the blind and E Washington Correctional Institute…they have stellar stats. But I will give that TT’s run defense has been superb so far this year.

10. Phenomenal Smith said:

    September 24th, 2008 at 9:32 am

    Our defense is a lot better than it was a year ago.

    As Winston Wolf would say, “Let’s not starting sucking each others’ dicks just yet.” UT’s opponents, all non-BCS conference teams, are a combined 3-8 on the season.

    UT’s defense may be better than last year – heck, may even be a lot better – but the stats above don’t reveal the difference. Let’s see what happens in conference play before you hoist Muschamp on your shoulders and tote him around DKR.

11. Bob in Houston said:

    September 24th, 2008 at 9:35 am

    Looks much better to me, HJ, even compared to last year at this point in the schedule, given the youth in the secondary, which is the focus of all pass-oriented offenses.

    YPP about the same, forcing punts better, RZ better.

    I still fear what Bradford and Co. may do, but things are different.

    I’d still be fine with 9-3, but I wouldn’t be depressed if it were worse.

12. TWC said:

    September 24th, 2008 at 9:40 am

    Marginal improvement seems like the best you can ask for with almost an entirely new defense, even if it is against chumps.

13. HenryJames said:

    September 24th, 2008 at 9:49 am

    We’re playing better with 2 freshmen and a sophomore starting in the secondary than we were a year ago playing with 3 seniors.

14. BrickHorn said:

    September 24th, 2008 at 10:00 am

    This is the new black. Texas is currently giving up 6.18 with Muschamp’s goal being 5.0

    I find that statistic and goal bewildering. If you give up 5 yards per pass attempt then, on average, you can expect to yield a first down every two plays. Since an offense gets four chances to gain 10 yards and a renewed set of downs, this goal seemingly does not bode well for stalling opposing offenses.

    Then again, there was only one team in the entire nation last year that is yielded fewer than 5 YPA (Ohio St. at 4.8). I suppose the statistical distribution of yards per attempt is somewhat binary, in that roughly half of the passing attempts will result in 0 yards (including interceptions) while the remainder will result in positive yardage (with a mean of nearly 10 yards). Assuming a team punts or kicks a field goal on 4th, there’s a good chance that a set of 3 straight passing attempts during a drive will lead to two 0-yard plays and a completion of less than 10 yards. There’s a better chance that 3 straight passes will lead to at least 10 yards (either through 2 passes of less than 10 yards or one of more), but you only need one 3-down stop in a drive.

15. Kafka said:

    September 24th, 2008 at 10:13 am

    kriess:

    The stats are not impossible. The QB’s tendencies are irrelevant. What is relevant (for example) is how long the QB has the ball before he is forced to get rid of it. Tedious but easy to measure. Valuable to know because if the number is small enough, the pass D can take a more aggressive approach.

    The coaching staff evaluates if the right defensive call was made when the video of the game is analyzed exhaustively after the game. Simple enough to do but probably not info that the coaches always want distributed widely.

    I’m pretty Mack would be interested in this stat, though.

    You state: “If you know the correct play to call against every O formation then you need to be on the sidelines with a headset.”.

    Could you point out what I wrote that indicated that I think I know the correct play to call against every O formation? Or was this just a straw man argument (i.e. a straw man argument is an informal fallacy based on misrepresentation of an opponent’s position)?

16. Bob in Houston said:

    September 24th, 2008 at 11:09 am

    “I find that statistic and goal bewildering. If you give up 5 yards per pass attempt then, on average, you can expect to yield a first down every two plays.”

    And this ultimately causes Mike Leach.

17. Huckleberry said:

    September 24th, 2008 at 11:17 am

    It sounds like BrickHorn is in favor of calculating the mean average yards gained per pass play called (including sacks), the standard deviation of yards gained per pass play called, and calculating the probability of three consecutive pass plays yielding 10 yards or more based on those figures.

    We could also do the same thing with called running plays.

    This would be the best way to determine whether a team’s rush defense or pass defense is really better. Same for the offense.

    Well, then you have to get into probability of longer plays meaning fewer first downs required before scoring a touchdown or field goal attempt. So you could eventually calculate the probability that an offensive possession would result in a touchdown based on different starting positions. Based on the percentage of pass versus run plays called, or for this exercise based on exclusively pass plays or run plays called.

    And now we start to see why stat freaks gravitate toward baseball. Much easier to analyze a game played in series than the ones played in parallel.

18. Bob in Houston said:

    September 24th, 2008 at 11:29 am

    While it looks like an all-pass offense would always work, even at 5 ypp a sack of more than five yards would “guarantee” that a first down could not be made, and that an incomplete pass on any of the first three downs would be much more likely to force a punt.

    But, if you were certain that a team never would run, how many rushers would you use? The answer obviously is not zero, but it would seem to me that two might be as possible as five or more. Just thinking on line here…

19. ponderos said:

    September 24th, 2008 at 11:33 am

    Texas is currently only giving up 11 points a game, much improved over last year’s 25.3 points per game.

    That’s about to change …

20. vvn8bs said:

    September 24th, 2008 at 2:57 pm

    INteresting post and thread.

    Another interesting thing is how context can really skew defensive stats. Total yards given up by a defense is greatly affected by pace the team’s offense’s pace of play.

    Also interesting…in all of the above stats, Mizzou and Texas are very close to one another…except scoring defense. The perception is that Mizzou’s defense has seriously underperformed while Texas’ has improved greatly. I tend to believe that, like batting average on balls in play in baseball, that both teams’ scoring defenses will regress to the mean in conference season.

21. AugieBall said:

    September 24th, 2008 at 5:48 pm

    Yards per attempt is the most underrated stat in football. People talk about rushing yards, YPC, and rushing TDs. Then they talk about passing yards, TDs, and INTs. Why isn’t yards per attempt in that group?

    Huckleberry,

    Or maybe stat lovers gravitate to baseball because they got depantsed by Westlake in football to the tune of 56-0. Food for thought.

22. BrickHorn said:

    September 24th, 2008 at 6:31 pm

    Another interesting thing is how context can really skew defensive stats. Total yards given up by a defense is greatly affected by pace the team’s offense’s pace of play.

    Yards per play or yards per drive are better stats, in my opinion. Although there is a possibility for truncation error due to the finite size of the field.

23. Bob in Houston said:

    September 25th, 2008 at 4:27 am

    I heard Mack talk about this last night. They also keep the stats differently than the NCAA would. If a QB looks to throw, and eventually runs, it goes in their pass stats. So, according to him, they’re slightly below the 5.0 right now.

24. Gene Claude said:

    September 25th, 2008 at 5:11 am

    I love the end of Brickhorn’s post saying “there is a possibility for truncation error due to the finite size of the field” and the next post from Bob starts with “I heard Mack talk about this last night.”

    Now that would be a really crappy press conference.

25. Kafka said:

    September 25th, 2008 at 5:18 am

    The offense prefers finite sized fields because it is much easier to score on a finite sized field than an infinite sized field.

26. Huckleberry said:

    September 25th, 2008 at 5:58 am

    That’s good to know, Bob, and I agree 100% with our staff’s way of looking at the stats if that’s the case. Plays should be analyzed based on what the offense’s intent was for the most part.

    AugieBall-

    Nice one.

27. BrickHorn said:

    September 25th, 2008 at 6:52 am

    The offense prefers finite sized fields because it is much easier to score on a finite sized field than an infinite sized field.

    Never mind what the offense likes. This finite field business is screwing with the stats.

28. Kafka said:

    September 25th, 2008 at 7:58 am

    If the yardage stats are kepts as reals instead of integers (which would be more accurate anyway), the problem goes away, right?

29. BrickHorn said:

    September 25th, 2008 at 8:43 am

    If the yardage stats are kepts as reals instead of integers (which would be more accurate anyway), the problem goes away, right?

    No. I don’t mean “truncation” in terms of round-off errors. I mean “truncation” in terms of certain plays being recorded as gaining fewer yards than they otherwise would have because the play resulted in a touchdown (i.e. it terminated in the end zone).

    For example, say the Longhorns go into the jumbo package and run it in from the 1 (yes, you have to really stretch your imagination here). That play is assigned 1 yard in the statistics. But how do we know that it would not have resulted in well more than 1 yard if it took place on, say, the 50? All touchdowns from the 1 are assigned the same statistical yardage value, whether the runner barely crossed the plane as his knee went down to the turf or went in untouched with nothing but open space in front of him.

    Or, look at it this way. Run the same touchdown play, with the exact same execution, from different parts of the field and you get vastly different statistical effects. A 99-yard touchdown run will jack up the YPP stat. The same run from the 1 will generally drop an offense’s average YPP. Yet it was (by assumption) the same play, with the same execution and the same result.

    That’s what I mean by “truncation error.” There is an inherent error in using average yardage stats to quantify the quality of an offense.

    Maybe a better stat would be expected value in terms of points or yards gained for drives that begin at some standard drive-beginning yardline (say, the 20). Offenses that can, in a typical drive, expect a higher number of points are superior to teams with a lower standard drive expected value (although points depends in large part on the quality of the kicking game, typically considered part of “special teams” and not “offense”).

    It would be possible to chart an offense’s expected value for drives starting in discrete yardage ranges over the full field. And the same stat could be applied to defenses, obviously with lower EV desirable in this case.

    Huck – do you want to crunch the numbers?

30. Huckleberry said:

    September 25th, 2008 at 9:24 am

    I actually ran that kind of analysis prior to the Rose Bowl against USC for each offense. I didn’t put that level of sophistication into it, I simply offered more stats than just total yards and yards per play. I calculated yards per possession as well as percentage of available yards gained. So a drive that started on your own twenty and ended at midfield would get you 37.5% of available yards gained. The obvious weakness with that, of course, is that a 10 yard touchdown drive after a turnover is valued as highly as a 90-yard touchdown drive after a good punt from the opponent.

    With my limited effort, though, I just rationalized that those kind of situations would wash each other out over the course of the season with the reporting of the total percentage of available yards gained for the entire season. That still leaves the problem of a team with a weaker defense requiring more yards to score, on average, than a team with a stronger defense (and special teams).

    What you’re calling truncation error is a very valid problem with analyzing offenses and defenses in football. The EV rating that you’re talking about really was my end goal when I started that analysis, but I never figured out a way to get to the final number because of the truncation error. How do I account for those 10 yard touchdown drives fairly when compared to a drive that goes from a team’s own 10 to the opponent’s 20 and yields a field goal? Intuitively, to me at least, the second drive is more impressive than the first, but it yields fewer points and a lesser percentage of available yards gained. So clearly total yards gained is still valuable.

    Determining the right way to handle those situations would be a major step forward. I don’t want to say it’s outside my reach, but it’s probably outside my reach given my available time to look at it. No doubt it would be an evolutionary step in analyzing offenses and defenses. But even that would face an insufficient data situation earlier in the year. By midseason it should be okay, I’d figure, with the added step of controlling for opponent strength included.

31. Bevosboss said:

    September 25th, 2008 at 10:01 am

    With a Sagarin SOS of #149, it is difficult to draw too many conclusions about the effectiveness of this defense. Evaluations will be more credible after Missouri.

32. BrickHorn said:

    September 25th, 2008 at 10:06 am

    Huck,

    The way to handle the problem you bring up, in my opinion, is to plot the EV or % yards gained versus starting drive position. Obviously, over the course of a season, teams will score more often on drives beginning past the 50 than drives that start at the 20.

    My thinking is this: you can gain a lot of information from looking at the EV (in points or percentage of available yards gained) versus starting drive position. You could compare two offenses this way, simply by comparing curves: the offense with the higher curve over the range is unquestionably superior.

    Some teams may cross over, with one team being better from far out and the other being better from the red zone. So, while you may not be able to say that one offense is clearly “better” than the other, the curve comparison gives you an understanding of why this is so.

    And, you can come up with all forms of ways to compare the expected score if one offense was plunked into the context of another team (i.e. given the other team’s defense and special teams). Basically, one team’s offense may have a very different average starting field position than another, due to its defense and special teams. To compare the effectiveness of that team’s offense to another’s, just calculate an expected points per drive by integrating the first team’s EV plot weighted by the second team’s expected field position.

    That’s probably confusing. Here’s an example. Assume Team A and Team B each have 5 drives in a half. You want to know whether Team A’s offense would have scored more points that Team B’s offense did in the same situation. The drives are as follows:

    Team A
    Starting / points
    20 / 0
    20 / 7
    45 / 7
    48 / 3
    42 / 0

    Team B

    20 / 3
    20 / 3
    20 / 3
    45 / 7
    43 / 7

    So, both teams started drives in the range 11-20 and 41-50 yards (and no others. Team A scored 17 points, and can expect to score 3.5 points from the 11-20 range and 3.33 points from the 41-50 yard range. Team B scored 23 points, and can expect to score 3 from the 11-20 and 7 from the 41-50 yard range.

    How much would Team B score if it was given Team A’s field position? The answer is the summation of expected value for each range times the number of drives Team A started in that range. In other words, 3×2 + 7×3 = 27 points. Team B would have done much better than Team A. And, Team A would have done worse than Team B given Team B’s field position (3.5×3 + 3.33×2 = 13.66).

33. Gene Claude said:

    September 25th, 2008 at 10:25 am

    I think net punting often has serious truncation errors. Anecdotally, 1/2 of Mizzou’s punts last year were of the pooch variety and suddenly many media outlets lamented MU’s punting difficulties based solely on gross or net numbers.

34. Huckleberry said:

    September 25th, 2008 at 10:32 am

    Brick -

    I agree with your methodology, at least for starters, and will look into it. Perhaps for this season I can publish it on a week-by-week basis for Texas and our upcoming opponent. So I’ll hopefully take a look at Arkansas tonight.

    This will, of course, be insufficient for a complete look because I won’t be able to account for opponents faced without running the same numbers for every team in the nation. But it should suffice for a first look.

35. BrickHorn said:

    September 25th, 2008 at 10:42 am

    Huck,

    Are you telling me that you don’t intend to track drive-by-drive statistics for every one of the 117 NCAA D1 football teams this season? Slacker.

    If I had a copy of MATLAB, I’d write some code to crunch the numbers. The real trick is finding and entering the data.

    One stat that would be an interesting grounds for comparison is EV for each team’s offense, given the national average starting field position distribution per game. This would (roughly) measure every team’s offensive performance if given the average special teams and defense.

    You could do the same thing for defenses, as well.

36. BrickHorn said:

    September 25th, 2008 at 10:43 am

    Gene – yes, punting is especially susceptible to truncation errors.

37. Huckleberry said:

    September 25th, 2008 at 10:48 am

    I have a copy of MATLAB on a CD somewhere. It’s probably a 10-year-old version. I don’t have it currently installed anywhere.

    But I can probably use Excel to do what I’m going to do.

38. BrickHorn said:

    September 25th, 2008 at 10:50 am

    Are you interested in selling it?

    It’s possible to use Excel, just tedious. MATLAB is more powerful.

39. HenryJames said:

    September 25th, 2008 at 10:54 am

    huckleberry@barkingcarnival.com

40. Vasherized said:

    September 25th, 2008 at 12:20 pm

    I just sold that prized e-mail address to a Korean spammer for $500 but he’s including a new copy of MATLAB. Thought it was a fair tradeoff.

41. Huckleberry said:

    September 25th, 2008 at 4:48 pm

    Here’s the output for the offensive EVs for this week.

    

    This early in the year there are some obvious data problems. Arkansas has only had two drives that have started on their opponent’s side of the field this year out of 35 possessions. So their 7.0 EVs in those areas is obviously misleading. The position where there’s the most data shows Texas with 13 possessions and Arkansas with 18 starting between their own 20 and 34 (inclusive). Texas’ EV is a little over twice that of Arkansas’.

    And while playing Alabama did hurt their EV in that range, they only managed 2.9 points per drive out of 12 possessions that began in that range against Western Illinois and Louisiana-Monroe.

42. Kafka said:

    September 25th, 2008 at 6:47 pm

    Brick:
    Now I understand what you are talking about when you say truncation.

    The quality of an offense is not only how many points it can score but also includes other factors such as: turnovers, ability to score quickly when necessary, ability to run clock when necessary, ability to score against the best Ds, ability to create good field position from bad field position, etc.

    It seems like a challenging problem. What makes it even more challenging is that the quality of the D being faced has to be factored into the computation.

    From a software prespective, I’m more interested in applications that provide a realtime assist for coaching decisions and applications that assist coaches in analyzing video.

43. Kafka said:

    September 25th, 2008 at 7:00 pm

    Brick:
    I don’t know how locked into matlab you are but I did a quick google on “open source math analysis package” and found a bunch of free stuff.

    Here is one link that looked interesting:
    http://math-blog.com/2007/06/02/3-awesome-free-math-programs/

    I have not used any of this stuff because I already have a copy of Matlab.

44. BrickHorn said:

    September 26th, 2008 at 6:31 am

    Interesting stuff, Huck. It appears that in terms of field position with a statistically significant sample size (5-50 yard lines), Texas has clearly been superior.

    Thanks for the link Kafka. I’ll check it out. Also, I agree that other factors may not be accounted for in an EV analysis like the one Huck did. Although you could tailor the same methodology to account for turnovers (assign them some negative value, especially in the case of a pick 6) or to compare an offense’s performance only against the top defenses it has played.

45. Atomic Teeth » Blog Archive » Muschamp and Defensive Stats that Matter (and Mizzou, of course) said:

    November 19th, 2008 at 7:04 pm

    [...] in September, Barking Carnival commented on “Defensive Stats that Matter,” as deemed by Muschamp (whoa, again with the HenryJames – he may need to get a TRO against me).  If [...]

Leave a Reply