College Football's Legs Race (Part 1)

College Football's Legs Race (Part 1)

Lance Roffers
Lance Roffers

Comments (75)

Great work Lance. Subjectivity is incredibly limiting and inevitably biased. I have to laugh every time somebody brags about tape. Well good for you. It is directly translated to, "I watched football players play football."

Likewise I get sick of moderators/administrators here listing Canes players who have advanced to the NFL from recent seasons, and trying to pretend it meant we were the equivalent of the upper tier college rosters. It has been simplistic and insulting for years, relying primarily on South Florida talent stereotype bias. I'm glad there is finally an analysis that smacks that in the face and effectively wipes it away.

I have followed the analytics for 7 or 8 years on certain draft sites, and most notably footballsfuture.com and also the Seahawks draft blogs. Seattle Seahawk bloggers are a cut above, largely because they picked up on tidbits from general manager John Schnieder and Pete Carroll years ago, describing certain traits and minimums they look for at specific positions. That led to Seahawks bloggers discovering the SPARQ numbers and how Seattle was applying them. Those bloggers have been able to identify one Seattle draft choice after another largely by focusing on what they look for. For example, Seattle never drafts cornerbacks with less than 32 inch arms. Seattle also places extreme emphasis on athleticism among running backs. They traded for Marshawn Lynch due to his domination of athletic testing. I could give one example after another.

Edge rusher has been the position that amateur evaluators have taken to a new level, in terms of predicting NFL success or failure. A poster named Waldo posted his breakthrough "Twitch" criteria on footballsfuture.com in 2011. Here is that link:

http://www.footballsfuture.com/phpBB2/viewtopic.php?t=439601

Waldo continued to update his formula for several years, along with the fits and rejects from each crop. Now he doesn't post anymore but others apply the criteria to each draft class. It has been maddening as a Dolphins fan to see first round picks like Dion Jordan and Charles Harris who were atop the "bust" projections based on Twitch, yet the Dolphins ignorantly take them anyway, and then can't imagine what went wrong.

Waldo's formula led to similar examinations from others, called SACKSeer and SURGE. I will not link to them but they are easily located via google. Every one of them has dominated subjectivity and traditional scouting methods.

I should point out that those Waldo-type formulas rely on more tests than apparently are available coming out of high school. The standing broad jump is used along with vertical jump to measure explosiveness. The 10 and 20 yard dash numbers are utilized and not merely the 40 yard number.

Perhaps the Canes can get a jump on matters by testing those numbers separately. Actually it probably isn't a jump. Other top schools have probably been doing that for years.

BTW, here is a guy who takes a different approach but it is also intriguing, IMO. He looks at scouting reports from past years and identifies which traits have aligned with NFL success. For example, at offensive line you need, "Plus Balance and Feet without weak anchor or demeanor OR Plus demeanor without weak power and either without weak punch or feet)"

https://forums.footballsfuture.com/...ictive-to-success-relative-to-draft-position/

That type of thing can be combined with the analytics. No sense in recruiting an offensive lineman who has balance issues or power issues or questions about demeanor.

Wide receiver has baffled me for decades. I am not surprised it does not align well with all the numbers. Seemingly it should be an easy evaluation but I have been wrong time and again. Finally I realized that those battles are equivalent to offensive lineman and hand fighting. You really need a mentally tough kid who can accept that he'll be open but the ball never arrives. He needs to accept that interference is not going to be called very often, even when warranted.

This guy James Cobern is the best at evaluating market share at each position and how it translates to NFL likelihood. He also places great emphasis on age, understanding that one year can be massive significance at those age brackets. He is taking a break right now but I'm sure he'll resume before the draft. He has a YouTube channel also.

https://draftcobern.wordpress.com/

Anyway, tremendous work. I look forward to future threads.
 
Great write up ---- WR's shocked me but started thinking about all of the bust we have brought in

PS lets get some DT's
 
Great work Lance.

Thanks so much for the response.

One thing that I probably didn’t make clear enough was that I am working to make projections to COLLEGE rather than the NFL.

I have done a lot of work with the NFL draft and am familiar with all of those places you listed. Even the metrics etc. Data has always fascinated me and I started with baseball way back in the early 90’s.

The data tells a different tale than the one it takes to succeed at the NFL level and if Manny can identify those traits that correlate to success with college prospects it will give Miami a decider edge.

Studies like this one aren’t trying to identify the freaks. Those guys are usually easy to see. This is identifying traits that have led to success in the past at the P5 level and can hopefully be used to filter and differentiate certain prospects in recruiting lists.

That’s why “one number” metrics fare do poorly at all levels- because they are tracking characteristics that don’t matter as much at each position. I want to find a way to emphasize what is truly important at each position and work to fine tune the model from there.

Hopefully that leads to greater success in identifying players but the HS landscape is even tougher and more vast than the NFL landscape.
 
Your analysis is exceptional @Lance Roffers! I've been fascinated with the results of statistical analysis for some time. The numbers can be very revealing. As you've mentioned, the challenge is determining what metrics to use.

Nor sure if you're familiar with QS9000 but it's a manufacturing quality system that I was involved in for some years that uses statistical analysis to measure quality performance. Do you use a particular program to assist in the calculations?
 
**** dude, do you have a job? I don't mean to come across like a jerk, it's just obvious you've put a lot of time into this and all the other stuff you post. Well done!
 
Advertisement
Recruiting is the lifeblood of any program. You can have all the X’s & O’s but you won’t win big without the Jimmy’s & Joe’s. Today, you will find numerous sites almost solely developed to the recruiting business. Fans of programs gather around their computer screens to see which prospect might be considering their school this week. In this series, I will take a look at the ever-growing “leg’s race” happening in college football and examine how data can help a program identify recruits.

A few years back I read an excellent article from the Harvard Sports Analytics Collective about the NFL Combine and the impact that athleticism testing had on the outcome of their NFL performance. It was a fascinating read and got me to thinking about the impact of athleticism testing in the college recruiting world. As such, I started on my journey towards a better understanding of how important athleticism is to a player transitioning into college football at the Power-5 level.

Methodology
To start you out, here’s a chart with positional results of athletic data from Harvard’s study to whet your appetite (correlation is reflected by color):
View attachment 77478
What I’m attempting to do is determine if there is a correlation in athletic testing results and All-Conference players at the Power-5 level. The athletic testing results come from verified Combine data such as Nike Opening, Under-Armour All-American Combine etc. The data includes regional events.

Therefore, my sample includes all players with verified testing results from these events from 2014-2018. Percentile rank shown for any player represents their percentile rank in SPARQ score against other individuals in their positional grouping. For the percentiles I only included individuals who completed all events and were able to calculate a SPARQ score from. Within any All-Conference percentages for each testing event I took all individuals who completed that event, so some individuals in those results might not have completed all events.

In addition to percentile rank for SPARQ scores at each position, I also calculated the number of standard deviations above/below the mean for each category and summed the totals. Here is a chart showing positional results from the Combine study with error bars included (bigger the bar, the wider the range):
View attachment 77479
I also wanted to test the results on individual teams’ recruiting classes. I chose the following teams as my sample: Alabama, LSU, Florida, Georgia, Florida State, Clemson, Ohio State, and USC. This sample group includes the elite teams (Alabama, Clemson), peer state teams (Florida, Florida State), a team from every area of the country (Ohio State and USC representing the Midwest and West, respectively), and a couple of programs that Miami recruits against often (LSU, Georgia).

The final area of data I included in my study was that of the ESPN300 recruiting lists and how they compared to predicting an All-Conference player versus simple athletic testing data. Recruiting lists have every resource available to them when making their list (camps, offers, athletic testing, production, interviews, film) versus just athletic testing data so the expectation would be that a list of 900-1200 players would perform much better than data-only.

Team Data
As you would expect, the team data results for Miami were not great and show just how far Miami has to go from a recruiting standpoint to catch up to the elite schools. I am especially happy to see that the results pass the “smell test” for the most part. The team results use the total of the standard deviations above/below the mean for each team rather than average SPARQ score. This way you are adjusting for positional differences in average SPARQ testing. For context, here are the average SPARQ scores for the positional groups for teams in our sample:
View attachment 77480

Here are the total number of Standard Deviations for the peer teams from 2014-2018:
View attachment 77481

Ohio State has recruited some supremely talented athletes during this timeframe and it shows up in these results. It’s interesting to note that Florida actually rates lower than Miami when it comes to recruiting athleticism. The data mirrors the W/L records of the teams fairly well, except for Clemson, who outperformed their athleticism recruiting by the greatest margin. Note: One of the limitations of the study is that not every player chooses to participate in athletic testing and cannot be included in the data.
View attachment 77483

Breaking it down by year also matches up fairly well with what you would expect to see from an athleticism testing standpoint, as Miami’s best recruiting class was in 2017. The worst recruiting class was in 2015. Here are the results with a yearly class rank for Miami among the nine teams. If Miami wants to improve on a national level, these results need to improve.
View attachment 77484

All-Conference Data
The driver of this study is to test for markers of athletic traits that continually popped up for players on All-Conference teams in the Power-5 leagues. The intention isn’t to say that if you have this trait you will definitely be an All-Conference player but rather to identify that others have consistently hit certain markers in order to identify recruits that have the highest percentage chance of success.

One observation that stands out is players who tested in the 90th percentile of a position group almost always become starting players at the P5 level and most become All-Conference of some type. Players in parenthesis represent a sampling of players who fit the criteria listed for each position.

QB- (Lamar Jackson, DeShaun Watson, Josh Rosen, Sam Darnold, Drew Lock, Jalen Hurts, Tate Martell, Jarren Williams)

It might surprise many of you to learn how athletic most All-Conference QB’s turn out to be in college. Players such as Sam Darnold and Josh Rosen are not known as athletic, dual-threat QB’s, but they tested out at very high levels. Here are the results of the QB data for P5 All-Conference Players:
View attachment 77485
The obvious correlation here is with foot quickness/speed. This makes a lot of sense if you think of how often a QB must manipulate the pocket in a tight space to keep a play alive. This information also joins very well with the NFL Combine data in weight, 40-yard dash, Short-shuttle having a large correlation to success.

RB- (Nick Chubb, JK Dobbins, AJ Dillon, Travis Homer, Bryce Love, Saquan Barkley, Lorenzo Lingard, Travis Etienne, Dalvin Cook, Joe Mixon)

View attachment 77486
For a position known for athleticism as much as RB, the data shows that there are merely minimum thresholds that must be met in order to profile as an All-Conference player. Having solid athleticism combined with excellent vision and strength seem to be the keys at this position.

WR- (Jeremiah Payton, Marquez Ezzard, Mike Harley)

View attachment 77487
The only position group of the entire study that didn’t show a high correlation between athletic testing and P5 All-Conference performance was the WR group. Much like the data regarding the NFL Combine showed little correlation, so to does the HS data. Route-running seems to be by far the biggest component to being a successful WR at any level. There was almost no correlation with height, weight, speed, SPARQ etc. Short-shuttle was the only metric that showed a positive and statistically relevant correlation which ties in heavily with route-running.

TE- (Hunter Bryant, Alize Jones, Brevin Jordan, OJ Howard, Brian Polendey, Chris Herndon, Dalton Schultz)

View attachment 77488
Much like WR’s, TE’s is the position that shows the second least correlation to athleticism testing, but there are a couple of clear markers to pass at this position, with minimum speed and strength performance required. It makes sense at a position like TE that fast and strong would be requirements. Weights ranged from 210 to 240 coming out of HS.

OT- (Mitch Hyatt, Jonah Williams, Kai-Leon Herbert, Martez Ivey, Mason Cole, Toa Lobendahn, Kolton Miller, Greg Little, Alex Leatherwood, Cam Robinson)
View attachment 77489
One position that showed heavily correlation with athleticism testing was at OT where being able to move and hit certain testing markers was imperative. For the most part, T’s that became All-Conference players were in the 260-pound range and added good weight over a period of a couple of years and had good athleticism. In fact, three of the T’s weighed less coming out of HS than recent Miami commit Zion Nelson.

OG- (Ross Pierschbacher, Braden Smith, Wyatt Teller, Sean Welsh, Saahdiq Charles)

View attachment 77490
Another position group where the testing markers match up very well with what you would expect of what asked of them on a football field. The G group is asked to go up against the biggest and strongest defensive players, so it is no surprise that the Powerball event matches up so well with excellent performance at the P5-level. Again, most All-Conference OL are not sloppy coming out of HS as the average weight of this group was 288 coming out of HS.

OC- (Michael Jordan, Brian Allen, Frank Ragnow)

View attachment 77491
You ask the C to anchor against a NT and to get to the second level with movement skills. Therefore, it makes sense that you want C’s who are quick, strong, and somewhat explosive. The results of the athletic testing measurables matches up perfectly with the position.

DE- (Lorenzo Carter, Derek Barnett, Tyquan Lewis, Austin Bryant, Chad Thomas, Carl Lawson, Jon Garvin, Josh Sweat, K’Lavon Chaisson, Xavier Thomas, Micah Parsons)

View attachment 77492
I did expect that SPARQ as an overall testing metric to correlate better at the position, but results showed more marginal correlation. What did show significant correlation though were common-sense factors such as strength, quickness, and explosiveness. For the most part they weighed more than expected, with only Duke Ejiofor of Wake Forest coming in at less than 215 pounds (he weighed 196 pounds). While the short-shuttle metric for 100% is fairly high at 4.87, the average was 4.58 and better reflects the type of numbers I would be looking for. The average 40-yard dash was only 4.98, so this metric really hits the mark of what you need as a minimum. I expected that number to be lower, honestly.

DT- (Solomon Thomas, Rashan Gary, Ed Oliver, Da’Shawn Hand, Taven Bryan, Gerald Willis, Raekwon Davis, Breeland Speaks, Jeffery Simmons, Harrison Phillips, Maurice Hurst, Vita Vea)

View attachment 77493
Perhaps the one position above all others, where if you post elite athleticism numbers, you are probably going to be a college star is DT. Only the Pac-12 (with two players) and Tim Settle (Virginia Tech) had even one below-average athlete at the position make All-Conference. The best course of action at this position seems to be to get 260-280 pound athletic freaks and let them gain weight and keep their athleticism.

LB- (Malik Jefferson, Dylan Moses, Raekwon McMillan, Dorian O’Daniel, Rashaan Evans, Jerome Baker, Devin White, Waynmon Steed, Roquan Smith, Micah Kaiser, Skai Moore)

View attachment 77494
Really, the big surprise for me here was the weight threshold. I expected a lot of players to be in the 180-pound range that were rangy in space and super athletic. The average weight coming out was 216 pounds, with only a couple even under 200 pounds. Much like WR seems to be dominated by route-running as a skill, LB seems to be dominated by an ability to use your instincts for production, but you have to have a certain level of speed and quickness to reach high levels.

CB- (Levonta Taylor, Adoree Jackson, Nigel Knott, Marco Wilson, Mark Fields, Grant Delpit, Trajan Bandy, Budda Baker, Al Blades Jr, Donte Jackson, Mecole Hardeman, Treon Harris, Mike Jackson, Bubba Bolden, Levi Wallace, Tre’Davious White)

View attachment 77495
At this position there appear to be two types: Either a freak speed athlete who is two or more standard deviations above the mean, or long and rangy CB’s with the arm-length to jam and play the ball away.

S- (Minkah Fitzpatrick, Derwin James, Jamal Adams, Quin Blanding, Justin Reid, Quincy Wilson, MJ Stewart, Tony Brown, Jaquan Johnson)

View attachment 77496
If anything, S’s actually posted more explosive numbers than CB’s, which surprised me.

ESPN300 Rankings
I wanted to have a means of comparison for the data metrics and I settled on the ESPN300 for a few reasons: It is easily accessible, and being the largest list it allowed for a larger sample size to smooth the comparison. For the time period 2014-2018, this allowed for as many as 1500 players possible to be All-Conference in their playing career at the P5 level. Note: I realize very few true freshmen make All-Conference at the P5 level so the number is probably much closer to 1200 than 1500.

I found that the vast majority (99.2%) of players on the list went to P5 schools, so the comparison would match P5 designation for All-Conference status. Seeing as how the list has access to combine performances, film reviews, production scores, All-State designations, offer lists (meaning Saban and Urban etc. can essentially do much of their evaluations for them), athletic testing data, and interviews (insight into their drive and personalities), you would expect the list to perform much better than athletic testing on its own.

Here are the percentages that the ESPN300 produced for the All-Conference lists:
View attachment 77497
These results are fairly intuitive with the top two positions being RB and DT. RB makes sense as it is the easiest position to transition from HS to college from an evaluation standpoint and DT due to the way athleticism translates to the next level at the position.

To compare the results to the athletic testing data I ran a Ridge Regression rather than a Least Squares Regression for the purposes of regularization. I won’t lengthen the article by going in-depth with the math stuff, but I can tell you that the correlation from simply going off of ESPN300 rankings for P5 All-Conference is around 0.45 and doing nothing other than paring down positions by the above athletic criteria gets you to around 0.38. Getting to 84% of the accuracy of a ranking list from websites that have access to infinitely more information and are able to rely on the best coaches in the country for backup evaluations is pretty impressive in my view. There is still a lot of noise in the data, as you would expect when you are dealing with the volume of high school players that are moving into the division I college level every year, along with injuries, scheme changes, transfers etc. but you are starting to see real value in evaluating athletic testing and the characteristics that make up each positional grouping.

As an example, by filtering the data results at the QB position to include only the players that fit the model characteristics listed above (< 5.0 40, <4.47 SS, 34’ Powerball, 190 pounds or more, >82 SPARQ) you get a nice group that includes out of 27 names:

Justin Fields, Josh Rosen, DeShaun Watson, Sam Darnold, Tua Tagovailoa, Jalen Hurts, Tate Martell, Jarren Williams, Jake Bentley, Sam Ehlinger, Brandon Wimbush, Adrian Martinez, Trevor Lawrence, Drew Lock, Lamar Jackson.

You’re starting from a pretty good place if that’s the type of list returned from simply filtering through athletic testing numbers. Taking that down to include only players from the 14, 15, 16 classes (to give time to establish as the QB) you are returned with 11 players out of 44 total QB’s who completed athletic testing:
View attachment 77498
That list returns a pretty nice group of QB’s and really allows you to limit your focus as a recruiter. Eight of the 11 QB’s have had pretty good college careers, with four of them already having been 1st round NFL draft picks. Will Grier missed the list by nine pounds or he would have been on there as well.

Of Note
I wanted to point out that when I started the endeavor, I almost didn’t push forward due to my belief that HS athletic testing would be so fluid once the athlete hits a college S & C program, I expected huge leaps in the testing performance once they reached the NFL Combine. The NFL Combine seems to be an excellent proxy to use since we are looking for P5 All-Conference potential and individuals who reach that status are generally invited to the Combine at the conclusion of their careers.

Since I also have an NFL Combine testing database available to me, I was able to easily cross-reference the players listed in both databases and found that- on average- players do not improve their athletic testing by massive amounts through college. There are certainly outliers who go to college, enter into a weight program and see their athleticism blow up (Myles Garrett immediately comes to mind), but by-and-large you are who you are as an athlete. In fact, several athletes regressed in their testing after gaining weight in college.

Next Steps
This article is part one of a three-part series. The next installment will review the current recruits in the ’19 class and compare each player to the criteria set for each position in order to determine if they meet certain thresholds for All-Conference potential as well as overall team data with the class. The third installment will review the overall roster and how each position group is positioned to return Miami to a national contender again, as well as where the deficiencies are from an athletic standpoint.


Great stuff. One small gripe - you can't add standard deviations. You can only add variances.
 
Great stuff. One small gripe - you can't add standard deviations. You can only add variances.

This is actually true. I was looking just to get a sum of things- and my correlation was on variance- but The Pythagorean approach is the way to get accurate sums of standard deviations.

Thank you.
 
Advertisement
all These teams playing with a different set of cards than the rest of us.
Even the suckeyes with all those freaks haven’t made the playoffs since their natty.
When cmr got here people said we were like 2-3 recruiting classes away.
So how far are we now?
 
all These teams playing with a different set of cards than the rest of us.
Even the suckeyes with all those freaks haven’t made the playoffs since their natty.
When cmr got here people said we were like 2-3 recruiting classes away.
So how far are we now?

We aren't far...Lost 2 coin flips

GOTTA GOTTA GOTTA recruit through the negativity though, have to hold on to this class
 
We aren't far...Lost 2 coin flips

GOTTA GOTTA GOTTA recruit through the negativity though, have to hold on to this class
I meant how far are we from losing or winning a coin flip to Clemson and the rest on that list.
 
Advertisement
Great work, Lance. In particular, I took note of your conclusion about the "they are who we thought they are" bit about how college athletes are largely what they were in HS (i.e, a cut above the rest, statistically).

Hopefully, this will put the final nail in the coffin of the "let's-recruit-a-bunch-of-2-star-diamonds-in-the rough-because-that-one-guy-that-one-time-became-a-first-rounder" crowd.
We used to get a lot more of those players and you would see the underrated emerge. There were many examples in the '80's and even '90's. I think the ratings back then we're done in the heads of a few retreads from other "professions" without resort to metrics, data, testing, etc. Examples were Max Emfinger (itinerant assistant coach and "scout"), Allen Wallace (ex-lawyer), Tom Le

Now HS ranking services are apparently relying on more testing and data, camp results, etc., so maybe their evaluations are better. So maybe stars are more accurate reflections of actual ability.
 
The stat that blew my mind is that 100% of All-Conference QBs ran below a 4.47 shuttle. Foot quickness is paramount.

I went back and looked at Malik Rosier since he came in as a dual-threat QB. Sure enough, he ran a 4.56 shuttle.

Jarren and Tate met the threshold.
BRAD KAAYA
 
Advertisement
Back
Top