I thought that myself when I read that article that talked about a Data Scientist(tm)
Right before the UConn game, I made some incendiary remarks about this coaching staff’s ability to coach. Clearly, my emotional state contributed to much hyperbole.
Now, with the standard disclaimers in place (the Big 10 season hasn’t begun, we started 4-0 last year, etc.), I offer an interesting development on Rich Rodriguez’s coaching ability.
In three years, RR has transformed an ungainly, Benny Hill offense into one of the best in the country. Some perspective on this is useful. He inherited offensive personnel as mismatched to his system as…well…words fail me:
The Borens, Malletts, Threets and Arringtons fled at the prospect of playground, Chinese fire drill ball that required a level of fitness beyond their experience and surely would destroy NFL aspirations like Taylor Lewan destroys donkeys. Three years later, it is safe to say that there is not a defense in the land that wouldn’t be gibbering in fear at the thought of playing Michigan this coming Saturday. Project yourself into the film room in Bloomington and imagine the discussion they are having.
Given his exploits at WVU, all of us wondered what RR would do with a higher level of talent. I think we are getting a sense of that. Yet, would anyone have said, after Tressel pulled in this class…
…and after Threetsheridammit debuted that RR would have passed Tressel two years later?
I touched upon the tantalizing potential of RR finding diamonds in the rough like Denard, Omameh and Odoms in my McBean series (to be continued in the off-season), and it appears that, both as a unit and player by player, Michigan has a superior offense to Ohio State. (For comparison, I am using Rivals Ratings [RR] for a finer comparison.)
Is the blossoming of the Michigan personnel heredity or environment? If the former, RR knows talent. If the latter, RR makes talent. If a combo, even better.
Some might say that more than one offensive line position would be up for debate. However, living in Ohio as I do, I had the pleasure of listening to some Buckeye postmortem radio show about three hours ago, and several callers complained that Ohio State's offensive line could not get a push against Eastern Michigan. Complaining about underperformance of offensive lines is an October ritual in Columbus. As a unit and especially considering backups, I wouldn't even look twice at Ohio State's offensive line, pedigree or no.
Quarterback? In my opinion only, I think Pryor is a very good quarterback with enormous potential to implode. Their backs? Eh. Their receivers; yes, a couple are very nice, but how can one be discontented with Roy and Clark Kent and the Billy goat?
What's the point of all this? Besides being stunned at how deep RR's reclamation project runs on the offensive side of the ball, it has not translated to the defensive side of the ball, which is why OSU would be a double digit favorite if we played THE game tomorrow. Our purpose is not the rehash the trail of tears that has been the defense, but to ask a more pointed question (and I ask sharik directly):
- How is it possible for a coach with such a sophisticated understanding of offensive schemes to not have an equally sophisticated understanding of the defenses most effective at stopping those schemes? In other words, can he do for our defense what he has already done for our offense, or is he a half of a coach?
Source Material: Original Post, Definitions, 2002 Class, Problems, 2002 and 2003 Classes.
The generally low level of activity on the last couple McBean posts is because the season is upon us. Or the posts aren’t very good. My ego and the advice of wolfman81 are sure it is the former; it appears this project is ideal for the off-season when football filler is welcome.
The plan, therefore, is to shelve for a few months the massive posts that rank the players class by class in exchange for amusing ourselves with surgically precise mgoboard posts that ask about individual players. Each recruiting class has a good handful of players on the bubble, and it is my goal to increase the survey participation beyond the current group of hard core McBean aficionados. This will take the form of an mgoboard post that has a short preamble, a link to source material and a survey about a single player.
My goal will be to get to 25 or 30 votes on each bubble player. If the mgoboard post falls off the front page before that total, then I will repost it at a later date. This should allow us to get to a statistically useful number of opinions about borderline players for when we finish the project in the off season.
Here is an example:
Preface: In August, I launched the McBean Rating System and asked the mgocommunity to help me rank every Michigan recruit at the end of their career as a point of comparison to their initial rating (using Rivals)*. Jake Long was a four-star recruit coming in and, after his career, he was a five-star going out. Kevin Grady was a five-star recruit coming in and will likely be a three-star going out. This subjective rating system depends heavily on definitions designed to maintain the same relative number of players in each Rivals rating bucket. The goal of the project: to develop a “collaborative, ongoing post-recruitment rating system that will allow us to determine if, in the Rich Rodriguez era, perfect-fit three-stars are more desirable than random four-stars.” In other words, to answer the Pat White question once and for all.
Player: David Harris
All American: No
All Big 10: Senior year
Drafted: 2nd round (47th)
Bubble Question: Was David Harris a four-star or a five-star player at the end of his career at Michigan? (Please review the definitions.)
Good idea? Bad idea? Additional information needed? Board? Diary? Do you think this will be productive?
*Since that time, I can no longer take ownership of this project as there have been significant contributions from several mgobloggers; this is a mgoblog community project now.
Note: The spectacular rollercoaster start to the season distracted me from our McBean effort. Back to work. Source Material: Original Post, Definitions, 2002 Class, Problems.
There was a flurry of concern about how our definitions, particularly the proliferation of McBean two-stars when we rarely recruit any Rivals two-stars, necessarily results in an overall decline in performance for any given class. This is a big deal when you recall that the purpose of this project is to allow us to judge whether a staff’s recruiting and talent development exceeds or falls short of expectations. We can say right now that they will almost always fall short of expectations; the question is how much. One day, we may adopt a variation of the proposed formula from wolfman81 to solve this problem (I say variation because I don’t think it goes far enough to compensate for the two-star problem – see the 2003 class RMS below, which narrows the gap as it should for a great class, but not enough…I think somehow weighting five-star players may be the answer):
Lastly, you asked me about my formula. It's really just the Root-Mean-Square. So add up for each player (star rating)^2. Then divide by total number of players. (This is the mean of the squares.) Now take a square root so that the numbers are comparable.
Example: Compare these 2 person classes (2 4 stars, vs. 1 5-star and 1 3 star)
2 **** -> Avg = 4.0, RMS = 4.0
1 ***** + 1 *** -> Avg = 4.0, RMS = 4.123
I'll ask the question this way. Would you prefer a class that is half 3-stars and half 4-stars (remember, I'm talking about McBeans here--so 12 All-Conference players and 12 servicable backups) or a class that is half 2-stars and half 5-stars (so we have 12 All-Americans in a single class and 12 guys who never play)? I know what my answer is (especially if we consistently recruit and develop that kind of talent).
I am rolling out this stat below the averages for your consideration.
In the end, we decided to finish out our McBean rating effort and go from there. With three classes in hand, we will be able to gauge the two-star problem and either wallow in that misery, as UMFootballCrazy wants, or create an algorithm to compensate the relative value of players, as wolfman81 wants. The Team Ranking analysis for the 2003 class demonstrates the two-star problem clearly…
…but we’re going to finish and circle back. I will probably even finish the 2005 class, even though we have active players.
Here are the first two final classes, 2002 and 2003:
For the 2003 class, Kraus was bumped to a four star, which was near unanimous except to UMFootballCrazy, who doesn’t like to ignore the NFL draft in this instance. I have felt bad for SanDiegoWolverine in the past because he keeps passionately arguing for certain guys and not getting his way – in this instance, he wins as both Rivas gets moved to a four-star (you can’t use the NFL draft as a tie-breaker for kickers/punters and he is a multi-year starter) and Richard gets reclassified as N/A.
JimHarbaughScramble is clearly still grappling with extreme Mundypobia, but sorry JHS, I can’t make a drafted DB a two-star no matter how many times you see this running through your mind:
Here is the preliminary 2004 class, which is our last class that we can call complete (the 2005 class has six active players). There are plenty of issues, and most of them seem to result from overrating our favorite players, like Henne and Hart.
There's a lot to debate in the 2004 class.
Before that, I wanted to give a summary of the informal voting that took place in the last thread as it related to the 2002 class (note: it would be helpful if more readers would vote – the number of votes was small, although the justifications were scholarly). Here was the 2002 class career ratings proposed after the McBean Rating definitions were debated:
The following votes were then cast:
Five-Star: Two votes
Four-Star: Three votes
Three-Star: Three votes
Two-Star: Zero votes
Three-Star: Two votes
Two-Star: One vote
Single votes for Bihl as a four-star, Breaston as a four, Tabb as a three and Riley as a three.
All votes are not equal as arguments of varying persuasiveness were offered. In the debate, wolfman81 almost perfectly captured what still remains a point of contention: how to use the NFL draft in our assessment -
Notice that in all of these cases, I consider the college career first. If there is some doubt about what category the player should fit into, only then do I consider the NFL draft. The NFL draft is a tiebreaker, but only when there is a tie to be broken. In my mind, 1st team All-American = 5-star McBean. All-Conference = 4-star McBean. Starter = 3-star McBean. There can be exceptions to this rule (Tom Brady was not All-Conference, but I think he should be 4 star due to his leadership on the field and other intangibles. As he was drafted, the NFL assessment backs this opinion up. The fact that he will likely be a first ballot NFL hall of famer cannot bump him up to the 5-star discussion; at the same time, it does not harm my 4-star rating.) But, I think that this rule sets a minimum standard.
The new 2002 McBean Ratings are as follows with only one change – Harris drops to a four-star. Mark Bihl was pushed hard as a four-star, but his career falls short of making a significant impact, just as Harris clearly falls short of an elite career.
The conclusion one would then draw from this data is that the 2002 class dramatically underperformed. To give you an idea, this would represent a drop in the team rankings (if the rankings were based upon average stars) from 10th to 21st, using an average of the 2002 through 2009 data.
Now, the problems:
- How do we handle the math for players like Dann O’Neill, Quinton McCoy and Taylor Hill? Players who never saw the field and did not stay. My original method counts these players in the Rivals list, but does not count them at all in the McBean list. For example with the 2002 class above, the 21 players in the Rivals list are all counted and the denominator is 21. For the McBean review of their careers, McCoy and Berishaj didn’t have a career at Michigan, so they are not counted and the denominator to determine the average star rating is 19. If we eliminate the two players from the Rivals list, do we fail to take into account the reference point of our evaluation: the quality of the entire incoming class as judged by a recruiting service? Yet, by taking them out of the McBean career list, do we fail to punish the staff for recruiting players who could generally be considered a bust? However, if we rate them a zero or even a two, we ensure that pretty much every class will underperform badly (counting both as a zero, the McBean average would drop to 2.81!). I am both befuddled and in need of a paragraph break.
- Second, the ease with which we rate invisible players as two stars devastates our average rating. We rarely recruit two-star players, yet because every team logically has many anonymous backups (seventeen preliminary McBean two-stars over the first five classes I’ve looked at), we have exceeded a ratings Chandrasekhar Limit and pulling each class down into a black hole of underperformance. Case in point, I offer the 2003 class which, if this class doesn’t qualify as a recruiting success and overperformer, then none will:
EDIT, August 29th, 9:24am: Gentlemen, after reading below and thinking about the problems presented, I feel we made a mistake with the definitions. Any viable Michigan players - this includes backups who do not see the field - should be get three stars. We could use this yardstick: any player who makes the two-deep during his career gets a third star. Also, we should be more generous with four stars for good starters like Bihl and Kraus.
Here's my reason: we must remain connected to our baseline system, Rivals. It is our point of comparison. Without a close connection, we can make no claims about under and overperformance. We recruit many three-stars, and while individually each kid hopes to be the next Pat White, the 30,000 foot reality is that those three-stars, en masse, are destined to be the backups. They are not underperforming, they are performing as expected.
This pushes some - not many and not enough to unbalance the finite four-star numbers - up a notch, because Mark Bihl and Adam Kraus cannot coexist with a three-star backup.
So, I am thinking we need to predominantly loosen our three-star standards to allow us to recognize backups as meeting expectations.
The McBean Definitions discussion was lively.
Only minor changes resulted from the debate:
- I removed the reference to a “five year starter.” Amazing that so many different sets of eyes (except cali4uofm) could miss that, although it was referenced elsewhere.
- I clarified the McBean four-star rating a little bit. The 2002 class has two guys who I initially rated as four stars – Mark Bihl and Rueben Riley – that I have dropped to three stars given that they were undrafted.
Three other issues were discussed:
- brad voiced a concern that if all starters at Michigan get three stars, there is no differentiation between a weak starter and a solid three-star guy like Chris Graham, for example. This is a reasonable observation, because I had Mark Bihl and Rueben Riley rated as four-star starters but backed them down to three (as I considered them borderline four-stars) because they were undrafted. So are both Mark Bihl and Darnell Hood three-star players? That needs to be further debated in this thread.
- There was some debate about punters and kickers, but I think the exception to a punter or kicker as a lower rated player can be handled in the rare event a Space Emperor decides to play ball on our planet.
- The dominant concern was using the NFL to assign career star ratings. SanDiegoWolverine voices this concern:
I don't think the NFL is that relevant in the sense that how our players perform in the NFL shouldn't change our perception of their value/production when they were at Michigan. I'd rather have Rod recruit players that dominate while they are at Michigan and underperform at the NFL than vice versa.
The draft is a national comparison, a national measure of the athletes.
And summarizes, I think correctly:
I agree with BlueBulls that ONLY the draft can inform a McBean rating, and only as a tie-breaker. Steve Breaston is a classic example. Four-star or five? He seems to be on the cusp – look at all the Michigan records and his current value to Arizona – but in the end, he is a four and the draft – 5th round – gives us a valuable assessment tool of his potential (developed at Michigan) at the time his career in college ended. So the NFL draft stays as a tie-breaker, and the final McBean Definitions will be:
"Elite" = 5*
"Significant Impact" = 4*
"Solid" = 3*
I think that you will find that, barring a few exceptions, most of those that get that fifth star will be first-rounders or high second-rounders; the second group will get drafted; and the third group will make up the bulk of the rest of the starters.
Let's begin with the 2002 class. There are some interesting borderline cases in this class. Given the currently assigned career ratings, this class underperformed significantly.
Note: On Monday, I posted Sylvester McMonkey McBean Helps Settle the Pat White Question Once and for All, complete with annoying HTML that I couldn’t edit out of existence without starting over. Won’t make that mistake again. What follows is the digested feedback for improving the McBean Rating System, the purpose of which, to quote myself, is to develop “a collaborative, ongoing post-recruitment rating system that will allow us to determine if, in the Rich Rodriguez era, perfect-fit three-stars are more desirable than random four-stars.” Our goal is to settle the Pat White Question once and for all.
Thanks all for the quality feedback. In the original post, I proposed that players be rated over their career at Michigan using definitions that approximated the spirit of the Rivals star ratings to see if, on average, players exceeded or fell short of expectations. The effort would span the later Carr years and, of course, the Rodriguez era going forward. How players performed relative to expectations would give the mgocommunity interesting information on the ability of a coach’s system to evaluate and develop talent. For more detail, the original post.
We began with these definitions.
These were found to be too loose. SanDiegoWolverine dabbled with the idea of a two-faceted rating system, but my IQ will not permit that level of complexity, so I used the following comments to revise the definitions above.
I'm not sure that there is a 1:1 correlation between the Rivals rankings and the McBean rankings.UMFootballCrazy found some numerical guidelines in Rivals:
Example (5 star): Elite, Dominant player, Complete package. Projects to the highest level. vs. Stud. Multi-year starter. All-something. NFL player.
I'll take "All-Something" to mean First team All conference in a BCS conference. That is 6 conferences x 22 players per all conference team = 132 -- plus kickers -- 5 star players. Does Rivals hand out 100 5 star ratings that often?
Players are also ranked on their quality with a star ranking. A five-star prospect is considered to be one of the nation's top 25-30 players, four star is a top 250-300 or so player, three-stars is a top 750 level player, two stars means the player is a mid-major prospect and one star means the player is not ranked.It seems that unless there is some loyalty to the proportion of star ratings allocated, then the McBean Rating loses its credibility. For example, if we dole out twice as many McBean five stars, is that Michigan developing talent or is that rose-colored glasses? It seems the NFL draft can serve in an advisory capacity, part of a checks and balances system.
UMFootballCrazy develops this:
5 Star *****He has more good stuff, but I steal so much of it down below, that I won't repeat it here.
This is a player who was drafted in the 1st round of the NFL entry draft and is thus regarded as one of the top 32 players in the nation (mirroring Rivals top 25-30 prospects) or baring that was an All-American and/or a 4 or five year starter (i.e. started as a freshman and started virtually every game they played at Michigan) and played at an "elite" level during their playing career.
Nedved963 cautions, however:
[M]oving forward with Rodriguez it's admitted his style is for college, and success in the Pros is not that great a correlation for the skill and effectiveness of Michigan's current players.I think Colin boils it down:
How you determine those expectations could certainly be done incorporating NFL performance.
And I think that is key. We don't want the NFL draft or an NFL career to answer entirely the question as to whether a player fell short of or exceeded expectations at the University of Michigan (i.e., Tom Brady), but it can be a tie-breaker, if you will, a check against overrating our beloved players (Breaston a five-star?) and ensuring proportion to the Rivals system.
As a result, I offer for final review, the new McBean Rating System definitions. I use the Rivals definitions (d'oh!) for the Rivals definitions. Special thanks to UMFootballCrazy as I steal a great deal of his language, but shift the emphasis, using the NFL as sort of a tie-breaker for those players on the cusp in the McBean ratings. However, I move away from UMFootballCrazy's proposed definition for the lower star players because I think there is a potential hazard to mishandling the lower rated players*:
Thanks again. If we settle on these definitions, I will re-rate the 2002 through 2006 classes.
* Again, to quote myself: "To be honest, I think some of the most important rating will take place down at the bottom of the scale. If you look at the classes above, what drags down a class (and gives Lloyd's classes a slight "underperform" grade) is guys ending up as two stars...being "invisible."
That is an area that has more impact on this analysis than whether Breaston gets a fifth. Since Michigan almost never recruits two star position players, is it fair that so many recruits end up as two stars because they can't get on the field?"