How can you say that Michigan has the most experienced units on either side of the ball?  Averaging over a junior status?  What depth chart are you looking at? 

Starting Defense:

• Roh - Soph
• Martin - Junior
• RVB - Junior
• Gordon - R-Freshman
• Demens - Soph
• Mouton - Senior
• Rogers - Senior
• Avery - Freshman
• Kovacs - R-Sophomore
• Vinopal - Freshman

Unless you're counting bench players or guys who we thought were starting (Woolfolk, Floyd, Banks, Ezeh) then something's broken.

Offense is a little closer because of two seniors on the line and some junior WR, but still surprised at the result.

Perhaps you should use median experience instead of average.  For example, a defense with 5 5th year seniors and 6 true freshman yields an average experience of 31/11 or 2.8181....  That's a shade under a junior average, but I think we all can agree that the example is not an experienced unit (nor is Michigan).

On the other hand, the example defense's median experience is 1.0, or true freshman, which is how I'd expect such a unit to perform.

It would be interesting to see how the numbers play out. 

I'd like to see your data weighted by star rankings of the players or some other method.  The idea being that if you've got a lot of 5th year 1 stars and walkon's, you're still not going to be very good, but a team with a handful of 5 star redshirt frehsmen and a couple seniors should do okay. 

Maybe you could color code the groups by conference or something. 

In other words, the effect of experience is probably real, but unmeasurable by your current method because other factors have a bigger influence.  You need to normalize the data somehow, and isolate the variable of interest.

has to be meaningful at some point, I wonder if games started would yield better results. I would think those stats would be available, and it shows actual game experience, vs, for example, a 5th year senior whose leadership and efforts can't be questioned, but who might lack the size, speed or talent to earn a starting spot. It might even account for the true 5 star talents and the Lewans of the world.

EDIT: Should be a reply to a post down below

Don't kid yourself, you did a nice analysis here...just because the numbers don't show anything doesn't mean we can't glean anything meaningful out of them. You're probably right, there are many variables to this that would change the analysis one way or another...what we do know is based on average/median years alone, there is no correlation. 

Others above have supported adding more or different variables, like games played rather than years. Or maybe a measure of "experience" worth trying is the ratio of  seniors to freshman on the field, as I would argue those two years are the most significant in determining experience or a lack thereof...or perhaps ratio of upperclassmen to lowerclassmen would be a better all-around variable for "experience". And maybe the inclusion of GURU star rankings as a factor in the analysis would yield more significant correlation (as we can surmise that having experienced 3-4 star players is usually way better than an experienced 2 star player)

Whether you delve deeper or not, thank you for the graphs and data and such. The stats nerd in me really wants to see this taken another step further, and there's quite a few possibilities with which to do it.

Players more often than not are better than the previous year.  I don't need a graph to tell me that defenses are better when more experienced.  They are.  End of story.

I think if you wanted to get a meaningful analysis.   Add up the amount of starts each player has had in their career.  Then get back to me.

Obviously Michigan was an outlier this year interms of stats.   What are the odds of having that many seniors on your defense that can't or didn't play.

Moundros, Ezeh, Rogers, Banks etc... 

You need to use a combination of career starts plus age to get a true "experience" factor that makes sense.  And you should not use the whole 2-deep.  Only the starting lineup, or maybe Starting lineup plus the 2 or 3 players that play more than 33% of the snaps on each side of the ball.  

There are tons of information sources out there aside from Rivals.

Mgoblue.com has TONS of info for one and espn.com for another.  

I have two viewpoints.

The first is that your definition of experience is really only describing how much opportunity the players had to go to the weight room at a University Level.  While this is important, I think it's also very common for all teams to require their players to lift weights.  There could be variations in quality of S&C, but that likely is strongly correlated to the coaching performance rather than just age.

Second, I was thinking that you are trying to show the correlation of experience to results, and to do that you need to isolate the talent variable.  If you weight the experience with a talent score as some have suggested, then I think all you'll do is prove that the most talented AND experienced teams have the best results on the field.  To which I say, "Duh".

Unfortunately to isolate, the only thing I can think of is that you have to look at all the teams in detail and put them into groupings based on their talent score, and then you can compare them across the variable of age or experience.

In the end, I think you are doomed.  This is impossible to prove because of the complexity of talent and experience and coaching.  Also Football is the most complex team sport, with 11 players on each side.

I remember when the Fab Five were at the Finals for the second time.  A friend of mine was a long time Tar Heels fan.  His opinion on Michigan was that they couldn't defeat North Carolina because they did not play as a team.  He thought Steve Fisher was an excellent recruiter (and of course in a few years we found out why), but was not a talented coach at all, and hadn't really done anything to develop the players.

In general I agreed, Michigan's performance frequently looked more like street ball than a college basketball game.  YET, that team of Five made it to the Finals twice before ever becoming upper classman.

Back to football, look at OSU, many times in the NC game, but many times coming out with a loss.  I don't think anyone would disagree that they clearly have the best talent pool of players in the Big Ten, they also pretty consistently put players on the field who have started for 2 years or more.

Now is that just because their players are older? no.  Is there talent pool proof of anything other than the ability of the coaches, maybe, recruiting is a different skill than teaching/coaching. But most of all if you try to analyze based on how many starts players have, then you are just going to prove that teams that find the best talent, and keep the talent pipeline flowing so they are always fielding a team of Juniors and Seniors, win the most games.

What you want to prove is that there is a team with incredible talent, who's coach has lead them to disaster this year (cough, Texas, cough) and see what you find.  Yes anecdotal instead of statistically cool, but such is the way with complex systems of too many variable.

We need to be careful about reading too much into the relative lack of correlation here.  All this says is that there appears to be no significant correlation between the 2 specific variables you have selected.  The failure to confirm this specific alternative hypothesis does not mean you have failed to reject the null hypothesis.

Alternative hypothesis: there is a significant correlation between average years of experience on a team's depth chart and its defensive FEI score. False.

Null hypothesis: there is no correlation between experience and performance on defense.  May still be rejected by confirming a different alternative hypothesis.

SE: VanBergen = 3 (3rd year getting PT)

N: Martin = 3

WE: Banks = 2

Sam: Roh = 2

Mike: Demens = 2

Will: Mouton = 4

Spur: C. Johnson or T. Gordon = 1

Bandit: Kovacs = 2

SC: Floyd = 2

WC: Rogers = 1

FS: C. Gordon or Vinopal = 1

Put the data in an ordered list and we get:

1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 4

mean = 2.09

median = 2

mode = 2

Any way you statistically look at it, it is a 2nd year defense in a somewhat new defensive scheme with new terminology.  Gee, why were we so bad?

Adding more variables to your model could help to explain more of the variance and improve your r-squared. Also, there could be confounding variables you're not considering. Years starting would be one I would look at right away. Also, if it were at all possible to consider their recruiting rating I would look at that, too (it might very well turn out that that isn't a useful predictor, but you never know until you try :) ).

And, as always with regression (and really most other parametric analyses), increasing sample size will almost always improve your p-value.

Overall, neat analysis. Definitely a lot of work put into this, and I applaud that.