Mean, mean regression

Here’s a piece from a post by Afghan Dawg over at Dawg Sports that caught my eye (for obvious reasons):

It is no secret that Senator Blutarsky is a fan of a metric known as “Regression to the Mean”.  Regression to the mean is the statistical way of saying that everything averages out overtime.  Things are never as good as they seem and they are never as bad as they seem.  If a variable is extreme on its first measurement, then it will be closer to its average on its second measurement.  It is likely that the first five years of Mark Richt’s tenure at UGA were one extreme, and the past five years have been the opposite extreme bringing us closer to our average.  Assuming we finish the season at 6-6, our mean with Mark Richt is 9 wins per season.

That’s not exactly right.  Here’s a wonky definition of the term, first of all:

Statistical tendency of a data series to gravitate towards the center of a distribution, provided it starts on the either end of the distribution and is free to fluctuate.

It’s the “free to fluctuate” part that’s key.  Regression to the mean is applicable to turnover margin because so much of that is dependent on random occurrences, such as fumble recoveries.  Sure, there’s some skill involved, but luck, for want of a better term, plays a more prominent role.

That’s not the case with a football program.  That’s not to say there isn’t a random element in play – what in life doesn’t have some randomness? – but there are many more controllable factors that work into the success of a program, such as resources, recruiting, scheduling, coaching quality, etc.  Indeed, that’s implicit in Afghan’s call for a new coach who can improve upon Richt’s record.  If wins were as random as turnover margin (and compare the fluctuation of the two in the Richt era to get a quick picture of that), wouldn’t changing coaches be little more than a crap shoot?


Filed under Georgia Football

26 responses to “Mean, mean regression

  1. Rick

    Another important observation is that the “long term” average when one only has 120 games of data is still significantly influenced by recent results, and so is also free to fluctuate (although within a smaller range).

    The program under Richt is, to this point, at it’s nadir. The long term average of 9 wins therefore is an average that looks worse than it has than at any point in his career (other than immediately after his first season). His “real” long term average is likely to be a good bit higher, unless he has simply become a worse coach relative to his peers (an explanation that obviously has a good deal of traction, but which reeks of recency effect).


  2. bigdawgfan

    Just hate how so many want a new coach with no sense of reason. Look what it has done for Notre Dame, Michigan, etc. Don’t give me Auburn, lets wait and see how the Tigers do the rest of the year with their one star Superman. I don’t want Georgia to be the next Notre Dame with a new coach every 4-5 years just because Richt is having a bad year or two.

    Do you think the Texas fans are calling for Mack Brown’s head? He has 4 losses this year. Are his good years behind him?

    Who do people think would be able to walk in and do a better job than Richt? Who out there is the next superstar coach who is dieing to be the new Georgia head coach? I just don’t see it.

    I think this team deserves one more year to pull it back together. If next year is a 7-5 year, then so be it, I will admit I was wrong.


  3. Scorpio Jones, III

    Of course changing coaches is a crap shoot….hopefully an educated crap shoot, but a roll of the dice still.

    For every Urban Meyer, or Mark Richt or Nick Saban, there is a…..well, the list of names is endless.

    There are just too many variables to make picking somebody to replace any established coach much more than an exercise in hope.

    There are Notre Dame fans, for instance and I know a couple, who actually thought bringing in their latest guy was going to immediately make the Domers relevant again.

    Hiring Dan Mullen appears, today, to have been a very good move for Missy State….but so did hiring Jackie Sherrill after a couple of years.

    Yep, it is a sho nuff crap shoot.


  4. 1 more year sounds right to me. Still I don’t expect the next HC to do better.
    2 & 8 against UF but the 11 years before that was 1 & 10(Dooley, Goff, & Donnan).
    Years ago the debate was who was the Dawgs biggest rival. I voted UF but others thought Auburn or Tech. Now if we upset Auburn we would be an unheard of 5 straight wins over them. A win against Tech puts us 9 & 1 under Richt against them. I suspect that a lot of fans are well satisfied with those numbers.
    Personally I would rather beat UF, then Tenn.,& then USCe than Auburn or Tech. Seems like we have done fairly well against all of them but the Gators. CMR is not gone yet.


  5. JBJ

    This is a silly analogy. What is your data series? Your performance consists of many things both random and planned. You can only control the things you practice and hope luck lands on your side sometimes. The fact is that you are coached on certain habits. It is the little things that win football games. The little things like RBs carrying the ball in the proper arm when running left or right to reduce the probability you will fumble. The little things like kicking the ball as far as possible on kickoffs and getting people downfield to tackle in order to improve your field position statistics. The little things like recognizing plays and putting yourself in the proper position. Little things like proper tackling technique. Little things like not fielding a punt inside the 10 yard line. Little things like not running backward on a punt return into the endzone during a tie game.

    The continued excuses for CMR and his inability to coach the little things astounds me. I will grant you that things are rarely ever as bad or as good as they seem, yet there is a tolerance level for poor performance that I have passed. Can CMR turn it around? Who knows. My point is that he should never have had to turn it around. A point that is lost on so many.


    • Mayor of Dawgtown

      Little things like falling on a fumble instead of trying to pick the ball up and run, then booting the ball.


    • AlphaDawg

      Like putting Smith in that situation to begin with? When our Faircatch Specialist has fielded those punts for the last 2 years.


    • Hackerdog

      So you believe that CMR has coached the kids to fumble, field punts inside the 10, not fall on fumbles, etc? Seriously? I think those things fall into the “failure to execute” column.

      I think you are criticizing Richt for Fabris, Martinez, and Jancek. But those guys are gone. Do you think the next head coach can fire them more than Richt already has? Seems pretty silly to me.


  6. Walt

    There has been a Mean Shift in the program. If the inputs have been the same the output should regress to the mean. I think the mean shift is this coaching staff has lost the ability to connect with the players to execute, ex. falling on fumble drills in practice, poor punt fielding, cannot defend the wheel route, all of which have been issues that alledgely the staff drills in practice.


    • Hackerdog

      And those are all skills that take time to manifest. Remember, Saban’s first Alabama team lost to LA Monroe. Should we hold Grantham to a higher standard than Saban?

      Most people would agree that Martinez & company failed to properly teach the fundamentals. But I’ve seen improvement this year. And I expected that it would take Grantham & his coaches time to undo the damage Martinez did and to instill new and better habits.

      Criticizing Richt for allowing Martinez to ruin the defense is OK, but the work to improve has begun. How would firing Richt now help that? Is there another HC/DC combination that can instantly instill perfect fundamentals and transorm a mediocre defense into the best in the land? I don’t think there is.


  7. 69Dawg

    What is being missed in the regression to the mean theory is that it is but a small part of the Laws of Probability. The old flipping a coin enough will get you a 50 50 result. Well it will but how many times have you flipped a coin and had it come up heads 5 times in a row. Every time UGA plays one of those “the opponent hasn’t won a game in 5 years” games I get sick at my stomach because the laws of probability say that they are due.


  8. Gen. Stoopnagle

    So, a guy who once claimed UGA as the equivalent of UAB doesn’t understand a concept from statistics? Big surprise.


  9. Jace Walden


    Afghan Dawg here. Just looking at the definition you provided, it specifically mentions a data series that is free to flucuate.

    I would argue that a win-loss record or number of Championships won are as free to fluctuate as the tendency to recover or not recover a fumble.

    There is a tremendous amount of skill and luck involved in both. I don’t think you can discount the luck involved in going on a 13-0 run any more than you can discount the skill and fundamentals involved in winning a turnover battle.

    The point I was trying to make is that Richt’s ability to coach isn’t as bad as a 6-6 season seems, but it’s also not as good as as his 13-1 season in 2002. The suggestion is that somehow, the recent failure is a reaction to his early success and that Richt, as much as we love him, really is a just an 8-9 win per year coach that sometimes does a little better and sometimes does a little worse. But he’s no Urban Meyer or Nick Saban as far as ability to win Championships and big games goes.


    • I would argue that a win-loss record or number of Championships won are as free to fluctuate as the tendency to recover or not recover a fumble.

      Statistically speaking, that’s not the case.


  10. Jace Walden

    Care to elaborate?


    • Go back and find my posts on regression to the mean as it relates to turnover margin. You’ll find (1) that fumble recovery in particular is about as random an event in college football as there is and (2) that there’s a much, much wider variance in turnover margin from year to year than there is in won-loss records.

      The other thing you’re doing differently that I did is that in analyzing the won-loss record, you’re comparing Richt to himself. With turnover margin, I was looking at bigger classes of data – either the SEC or D-1 football.


      • Mayor of Dawgtown

        While I agree with you generally on your theory about this, Senator, there remains a problem. If, for example one team does not try to get a fumble at all, that would distort the probabilities and hence “regression to the mean” would not work. Similarly, if a team had been so badly coached so as to not understand that the first thing is to gain possession, then try to advance the ball, and consequently failed to get its fair share of fumble recoveries because of ineptness, everything would then be completely out the window, probability-wise.


  11. Bulldog Joe

    I think I’ve wandered over to The Hive.