Writers of Pro Football Prospectus 2008

Most Recent FO Features

MurrayLat13.jpg

» Scramble for the Ball: With All the Fixings

An idiot's (two idiots'?) guide to Thanksgiving football, prepped and primed for the monsters-in-law who only watch these three games in a year.

17 Mar 2016

Tight Ends and the Combine

Guest column by Phil Watkins

While there is some debate about the importance of combine performance, there is certainly little doubt that some NFL positions require the athleticism that the combine attempts to measure. The success of crossover athletes like Antonio Gates and his successors certainly provides some anecdotal evidence that sizeable and exceptional athletes tend to excel at the tight end position. Indeed, it seems reasonable to hypothesize that any elite tight end would need to be an exceptional athlete to both block on early downs and catch passes in latter ones. To help explore this hypothesis, a sample of all combine data since 1999 was employed to model the probability that a given tight end would earn a legitimate (as in, not an injury replacement) selection to a Pro Bowl. While such a biased outcome has some limitations, the use of a binary (yes/no) measure to define NFL success is mathematically convenient.

The inherent trouble in comparing NFL hopefuls using combine data is that heavier men are at a natural disadvantage: Newton's second law of motion says that acceleration is proportional to force and inversely related to mass. Simply put, this means that a heavy tight end must apply a greater amount of force to achieve the same acceleration as a slimmer wide receiver. In a sport that revolves around moving other men against their will, mass adjustment should be the first step in attempts at meaningful analysis, particularly for a hybrid position like tight end. Football Outsiders' speed score for running backs has attempted just this. To this end, some additional metrics were computed that are theoretically proportional to physical quantities for all tight ends participating in a combine between 1999 and 2011, meaning they have each had at least five seasons in the NFL, ample time to earn a Pro Bowl selection:

  • Adjusted Vertical (High-Point): Vertical Jump + Height
  • Adjusted 40-Yard Dash (Momentum): Weight / 40 Time
  • Adjusted Broad Jump (Energy): Weight · Broad Jump
  • Adjusted Cone Time (Agility): Height / Cone Time
  • Adjusted Shuttle Time (Direct): Weight / Shuttle Time

As there were only 12 legitimate Pro Bowl selections between 2000 and 2014, pro day data was used to prevent missing values in that smaller sample. Exploratory analysis of the five composite statistics above suggested that the first four factors were at least marginally significantly different (p<0.10) between Pro Bowlers and non-Pro Bowlers. This left the following summary statistics to consider:


Tight End Combine Stats, 1999-2011

High-Pt
(inches)
Momentum
(ft•lbs/s)
Energy
(ft•lbs)
Agility
(inch/sec)
Bench
(reps)
Min 101.0 5724 2050 9.2 7
Q1 106.9 6204 2326 10.3 18
Median 109.0 6432 2408 10.6 20
Q3 111.5 6588 2510 10.9 23
Max 118.0 6966 2768 11.6 34

One doesn't need to be a physicist to see that the units for adjusted dash time are proportional to kilogram-meters per second, a unit of momentum. Similarly, adjusted broad jump is measured in foot-pounds, which is proportional to Newton-meters (Joules), a unit of energy. Adjusted Vertical is essentially the "high-point" potential of the player from a standing position: how far above the ground they can elevate. While agility (adjusted cone) doesn't have a convenient physical unit, it was used based on the understanding that taller players should have a harder time altering the direction of their higher center of gravity. Bench press reps were left unadjusted, as all players lift 225 pounds as many times as possible.

While there's certainly no magic formula for predicting tight end success in the NFL based on the above factors, one can apply statistical methods and models that give rise to odds and probabilities. One such method is known as logistical regression, which is nothing more than standard linear regression applied to the odds of success, which happen to be log-linear in nature (i.e., linear after a log transformation). However, correlation analysis uncovered a minor speed bump in modeling: weight-adjusted broad jump was strongly correlated with both high-point potential (r=0.54) and weight-adjusted 40-yard dash time (r=0.73), so this factor may not be used in conjunction with the other two factors when modeling. Electing to ignore adjusted broad jump, three iterative search methods (forward, backward, stepwise selection at p<0.10) of the remaining four variables (High-point, Momentum, Agility, Bench Press) suggested that only adjusted 40-yard dash (p=0.008) and high-point potential (p=0.010) were sufficiently significant factors in Pro Bowl selection. (Aside: the search methods included all interaction terms, so we can rest assured that these factors are additive in their contribution to the log odds rather than multiplicative.) This finding doesn't mean that bench press or agility are unimportant, only that these factors do not add enough information to justify their inclusion in this statistical model. That leaves us with the following equation:


Pro Bowl probability=
e^((Adj.40•0.532) + (HighPt*0.254) - 59.9)
--------------------------------------------------------
1+ (e^((Adj.40•0.532) + (HighPt*0.254) - 59.9))

Here's what this formula says about tight ends who have participated in the combine between 1999 and 2011 and been selected for at least one Pro Bowl:


Pro Bowl Tight Ends Who Participated in NFL Scouting Combine from 1999-2011
Player Name Height Weight 40 Time Vertical High-Pt Adj. 40 Prob (PB)
Vernon Davis 75 254 4.4 42.0 117.0 58.0 66.2%
Jimmy Graham 78 260 4.5 38.5 116.5 57.4 55.6%
Jordan Cameron 77 254 4.5 37.5 114.5 56.1 27.4%
Greg Olsen 78 254 4.5 35.5 113.5 56.3 24.6%
Marcedes Lewis 79 261 4.8 37.0 116.0 54.4 18.3%
Dallas Clark 76 257 4.7 37.5 113.5 55.3 16.1%
Jason Witten 78 264 4.7 31.0 109.0 56.8 12.0%
Rob Gronkowski* 78 264 4.7 33.5 111.0 55.1 8.4%
Alge Crumpler 75 266 4.8 34.0 109.0 55.9 7.8%
Julius Thomas 77 246 4.6 35.5 112.5 53.0 4.2%
Todd Heap 77 252 4.7 32.0 109.0 53.8 2.7%
Chris Cooley 76 262 4.9 33.0 109.0 53.8 2.7%
*Pro Day values used

With the exception of Rob Gronkowski (who was battling a back injury prior to his pro day), these rankings of size-adjusted athleticism seem about right. Also, we must remember that correlation is not necessarily causation. With all the recent success of NFL tight ends with collegiate basketball backgrounds, it's practically impossible to separate individuals with basketball skills from those sizeable individuals who can merely jump high and run fast. Regardless of the limitations, one should always realize that "while all models are wrong, some models are useful." With this in mind, let's see how we might translate these estimated probabilities into a useful rule for forecasting the success of a tight end given his athletic profile based on high-point potential and momentum score alone. A histogram of the estimated probabilities of our retrospective cohorts (1999-2011 combine tight ends) seems a good place to start:

Evidently, there's a steep drop-off in the number of players with Pro Bowl probability of 5 percent based on the derived model. Indeed, closer inspection reveals that a 6 percent probability is the ideal cut-point to balance the true positive and false negative rates:


1999-2011 Pro Bowl
Tight Ends
Non-Pro Bowl
Tight Ends
P(PB) > 0.06 9 46
P(PB) ≤ 0.06 3 142
True Positive: 16.4%
False Negative: 2.1%

With 46 forecasted individuals not yet achieving a Pro Bowl, it seems worthwhile to examine the list of hopefuls (or short-comers) up to 2011:


Unfulfilled Potential: Pro Bowl Candidates, 1999-2011
Name Height Weight 40 Time Vertical High-Pt Momentum Prob (PB)
Brad Cottam 80 270 4.63 33.0 113.0 58.3 45.6%
Leonard Pope 80 258 4.62 37.5 117.5 55.8 41.4%
Matt Jones 78 242 4.37 39.5 117.5 55.4 35.5%
Jared Cook 77 246 4.49 41.0 118.0 54.8 31.4%
Virgil Green 75 249 4.54 42.5 117.5 54.8 29.3%
Schuylar Oordt 78 261 4.63 36.0 114.0 56.4 27.8%
Zach Hilton 80 267 4.83 35.0 115.0 55.3 21.7%
Rob Housler 77 248 4.46 37.0 114.0 55.6 20.4%
Chris Luzar 79 270 4.76 32.5 111.5 56.7 19.7%
John Owens 75 265 4.77 38.5 113.5 55.6 18.0%
Anthony Becht 78 270 4.78 33.5 111.5 56.5 17.8%
Ben Watson 76 258 4.57 35.5 111.5 56.5 17.6%
Visanthe Shiancoe 77 251 4.65 39.5 116.5 54.0 16.9%
Joe Klopfenstein 78 255 4.63 36.0 114.0 55.1 16.2%
Jermaine Gresham 77 261 4.66 35.0 112.0 56.0 16.0%
Name Height Weight 40 Time Vertical High-Pt Momentum Prob (PB)
Quinn Sypniewski 79 268 4.77 32.5 111.5 56.2 15.6%
L.J. Smith 75 258 4.62 37.0 112.0 55.8 14.9%
Robert Johnson 78 278 4.96 33.5 111.5 56.0 14.7%
Larry Brown 77 278 4.94 34.0 111.0 56.3 14.6%
John Lumpkin 79 274 4.98 34.5 113.5 55.0 14.2%
Tony Scheffler 78 254 4.54 33.5 111.5 55.9 14.0%
Adam Bergen 77 259 4.78 38.0 115.0 54.2 13.4%
Jim Kleinsasser 75 272 4.79 34.5 109.5 56.8 13.3%
T.J. Williams 75 269 4.72 34.0 109.0 57.0 13.1%
Alphonso Collins 74 281 4.83 32.5 106.5 58.2 13.0%
Doug Jolley 76 251 4.6 38.0 114.0 54.6 12.8%
Martellus Bennett 78 259 4.68 34.0 112.0 55.3 11.8%
Weslye Saunders 77 270 4.82 33.5 110.5 56.0 11.6%
Anthony McCoy 77 259 4.71 35.5 112.5 55.0 11.2%
Mike Pinkard 77 259 4.69 35.0 112.0 55.2 11.2%
Name Height Weight 40 Time Vertical High-Pt Momentum Prob (PB)
Michael Allan 78 255 4.71 36.0 114.0 54.1 10.5%
Joel Dreessen 76 260 4.72 36.0 112.0 55.1 10.5%
Scott Chandler 79 270 4.78 30.0 109.0 56.5 10.3%
Cuncho Brown 76 272 4.78 31.5 107.5 56.9 8.9%
John Jones 76 248 4.63 38.5 114.5 53.6 8.9%
Dan Campbell 77 263 4.85 36.0 113.0 54.2 8.7%
Kellen Davis 79 262 4.59 28.0 107.0 57.1 8.7%
Kelly Griffeth 77 284 4.9 28.0 105.0 58.0 8.4%
Dan Curley 76 254 4.63 35.5 111.5 54.9 8.4%
Brandon Pettigrew 78 263 4.8 33.0 111.0 54.8 7.2%
Ben Hartsock 76 263 4.8 35.0 111.0 54.8 7.2%
Bennie Joppru 76 272 4.78 30.5 106.5 56.9 7.1%
Nate Lawrie 78 264 4.87 34.0 112.0 54.2 6.8%
David Thomas 75 252 4.67 37.5 112.5 54.0 6.8%
Luke Stocker 77 258 4.68 33.0 110.0 55.1 6.7%
Dorin Dickerson 74 226 4.4 43.5 117.5 51.4 6.1%
Includes all tight ends who participated in NFL scouting combine between 1999 and 2011 with at least 6% chance of making a Pro Bowl, but not (or not yet) named to a Pro Bowl roster.

A few big names are above, with some Pro Bowl alternates and replacement players as well. For those of you critical of the Pro Bowl status outcome, we might also compare the average Approximate Value (thank you, Pro-Football-Reference) of our two groups:

Still, as our method is based on historical records, it is critically important to remember that correlation is not causation, and test this rule in subsequent data. While the cohort of tight ends who have participated in combines since 2012 may be too inexperienced to expect meaningful Approximate Value comparison, with the recent selections of Travis Kelce (12.8 percent Pro Bowl probability) and Tyler Eifert (6.6 percent), we do see that this rule seems to predicts Pro Bowl selection a bit better than it did in the training set (though not quite "statistically significantly" at p=0.089 due to the small sample of only two years' time):


2012-2013 Pro Bowl
Tight Ends
Non-Pro Bowl
Tight Ends
P(PB) > 0.06 2 7
P(PB) ≤ 0.06 0 20
True Positive: 22.2%
False Negative: 0.0%

Though only two tight ends who have been in the combine since 2012 have been selected to the Pro Bowl, several others seem like promising candidates based on their combine performances. Here's a look at all tight ends who participated in the combine since 2012 with at least a 6 percent chance of Pro Bowl selection, including the two who have made it in Kelce and Eifert:


Top Pro Bowl Candidates, 2012-2015 Tight Ends
Player Name Year Height Weight 40-Yard Dash Vertical Jump High-Point Adj. 40 Pro Bowl Probability
James Hanna 2012 76 252 4.43 36.0 112.0 56.9 23.3%
Nic Jacobs 2014 77 269 4.76 35.0 112.0 56.5 20.0%
Michael Egnew 2012 77 252 4.52 36.0 113.0 55.8 17.7%
Jesse James 2015 79 261 4.83 37.5 116.5 54.0 17.3%
Joseph Fauria 2013 79 259 4.72 35.5 114.5 54.9 16.4%
Nick Kasa 2013 78 269 4.71 31.5 109.5 57.1 15.4%
Vance McDonald 2013 76 267 4.69 33.5 109.5 56.9 14.2%
Travis Kelce 2013 77 257 4.63 35.0 112.0 55.5 12.8%
Dion Sims 2013 77 262 4.75 35.0 112.0 55.2 10.8%
Jace Amaro 2014 77 265 4.74 33.0 110.0 55.9 9.8%
Troy Niklas 2014 78 270 4.84 32.0 110.0 55.8 9.3%
MyCole Pruitt 2015 74 251 4.58 38.0 112.0 54.8 9.1%
Chris Gragg 2013 75 244 4.50 37.5 112.5 54.2 7.7%
C.J. Fiedorowicz 2014 78 265 4.76 31.0 109.0 55.7 6.9%
Tyler Eifert 2013 78 250 4.68 35.5 113.5 53.4 6.6%
A.C. Leonard 2014 74 252 4.50 34.0 108.0 56.0 6.4%

Onlly nine tight ends completed both the 40-yard dash and the vertical jump events in this year's combine, and by this valuation method, none profile even close to that athletic pro Bowl caliber tight end; Beau Sandland has the best odds so far at just 3.5%. Of course, we'll need to wait and see how pro day numbers change our findings, but this method appears consistent with common perception in concluding that this year's tight end class is a weak one.


2016 Tight End Prospects
Player Name College Height Weight 40-Yard Dash Vertical Jump High-Point Adj. 40 Pro Bowl Probability
Adams, Jerell S. Carolina 77 247 4.64 32.5 109.5 53.2 2.3%
Anderson, Stephen California 75 230 - 38.0 113.0 - -
Braunecker, Ben Harvard 75 250 4.73 35.5 110.5 52.9 2.4%
Duarte, Thomas UCLA 74 231 4.72 33.5 107.5 48.9 0.1%
Grinnage, David N.C. State 77 248 4.90 29.5 106.5 50.6 0.3%
Hemingway, Temarrick S. Carolina St. 77 244 4.71 30.5 107.5 51.8 0.6%
Henry, Hunter Arkansas 77 250 - - - - -
Higbee, Tyler W. Kentucky 78 249 - - - - -
Hooper, Austin Stanford 76 254 4.72 33.0 109.0 53.8 2.7%
Malleck, Ryan Virginia Tech 76 247 - 34.5 110.5 - -
McGee, Jake Florida 77 250 - - - - -
Morgan, David Texas-San Antonio 76 262 5.02 30.0 106.0 52.2 0.5%
Sandland, Beau Montana St. 76 253 4.74 35.0 111.0 53.4 3.5%
Vannett, Nick Ohio St. 78 257 - 30.5 108.5 - -
Williams, Bryce East Carolina 78 257 4.94 29.5 107.5 52.0 0.7%

In summary, there is reasonable evidence that using height-adjusted vertical jump and weight-adjusted 40-yard dash times can divide the lengthy list of tight end combine participants into those with a potential pro Bowl ceiling (about 25 percent of all prospects) and those who may fall short (75 percent). Analytical methods such as these can be a massive time-saver when it comes to making decisions regarding these potential recruits, but we must understand that reducing an individual to a single number cannot perfectly reflect all available information. For example, while the beastly Jared Cook (estimated 31.4 percent chance of a Pro Bowl) looks great on paper, his 50ish percent catch rate in college (71-of-140 in games with play-by-play data) should raise some red flags of his Pro Bowl potential. Furthermore, a specimen like Vance McDonald (14.2 percent Pro Bowl likelihood), while worth his second-round pedigree, may not blossom until a freakish athlete like Vernon Davis (66.2 percent Pro Bowl likelihood) has vacated the depth chart. Finally, a player like Michael Egnew (17.7 percent) who operated in a spread offense in college may lack the blocking proficiency required for the NFL. One must always temper statistical analysis with common sense, and balance such information with other factors that may be less quantifiable.

You can follow Phil Watkins on Twitter at @Advantalytics.

Posted by: Guest on 17 Mar 2016

13 comments, Last at 01 Jul 2016, 3:34pm by logan35

Comments

1
by pads_of_the_hands :: Thu, 03/17/2016 - 11:52pm

Engaging read. An a topic of interest for me, as well; how measurable athleticism manifests into on-field performance. I've been slowly crafting an analysis of how these variables interact in game-play and failed to consider adjusting the combine measurables with sound theory as was done in this article. Also, good job communicating the concepts of logistic regression. Well done.

The combine data I have been using has inconsistent coverage of wingspan but I am curious how that would factor in your analysis. I suspect it might be redundant with the adjusted vertical data point.

2
by Megamanic :: Fri, 03/18/2016 - 3:20am

LaQuan McGowan's adjusted 40 is nearly 74 - how many pro-bowls does that project to? ;)

4
by TheSportistician :: Fri, 03/18/2016 - 9:35am

LaQuan McGowan participated as an Offensive Lineman in the Combine. As such, it would be extrapolating (operating outside the data we used) to attempt to apply this method or formula to him to predict if we expect him to make one or more pro-bowls. Still, Jason Peters had a very similar story and ended up making multiple pro-bowl appearances as an offensive lineman, so it's possible that a similar method (likely yielding different factors/formula) could be applied to offensive linemen as well.

3
by Aaron Brooks Go... :: Fri, 03/18/2016 - 8:45am

"Similarly, adjusted broad jump is measured in foot-pounds, which is proportional to Newton-meters (Joules), a unit of energy."

It's also a unit of work and torque, which is just as legitimate an interpretation. Be careful generalizing combine tests into fundamental units. Fuel efficiency can be measured in acres. Radians, g's, and Smoots are all measured in the same units.

5
by sharky19 :: Fri, 03/18/2016 - 11:43pm

I'm curious how Clive Walford graded out through this metric, he was pretty impressive last year.

6
by TheSportistician :: Sat, 03/19/2016 - 3:55pm

Working out this request as an example:

HighPt = Height + Vert = 76 + 35 = 111

Adj.40 = Weight / Dash = 251/4.79 = 52.4

Plug these into this formula:

e^((52.4•0.532) + (111*0.254) - 59.9)
--------------------------------------------- = 0.021=2.1%
1+ (e^((52.4•0.532) + (111*0.254) - 59.9))

If the formula is annoyingly complex, just use this lookup table for an approximation:

http://sportistician.blogspot.com/2015/06/retrospective-study-of-nfl-tig...

7
by Will Allen :: Mon, 03/21/2016 - 9:53am

Two "unfulfilled potential" candidates caught my eye, being Vikings who were on the same roster for 5 years. One of them had few years with stats that might catch the eye, since he was primarily a pass catcher. The problem for Visanthe Shiancoe is that he was not a good one, despite having some years with decently large reception numbers, and he was below average as a blocker. He had terrible ball skills, so he could gain some seperation but frequently could not finish the play, and was never even an average blocker. The Vikings did not get good value from that free agent signing.

The other guy was Jim Kleinsasser. who played 13 years with the Vikings, and was never a good receiver (although he was becoming decent prior to injuring his knee in 2004, in his sixth year), but the reason why the Vikings had him on the roster for 13 years was because he was a TREMENDOUS blocker, versatile and consistent. He was tight end who could handle speed rushers one on one, which most offensive coordinators try to do with tight ends at their team's peril, and in the run game, when you told Kleinsasser to set the edge, By Munoz, that was an edge that was set, even against much larger defensive ends. He delivered terrific value to the Vikings.

Evaluating football player performance quantitatively can be really, really, hard.

8
by tuluse :: Mon, 03/21/2016 - 1:57pm

I want so badly for the NFL to start tracking when players in the game like the NBA does. It would open up so many opportunities like what is the Vikings rushing DVOA with and without Kleinsasser.

9
by Aaron Brooks Go... :: Tue, 03/22/2016 - 1:34pm

Kleinsasser jumped out at me, too. The analysis, useful as it is, is completely oriented towards the type of TE who is really a tall, slow WR. It entirely ignores the genre of TEs who spend half of their career as utility FBs.

But that requires being able to usefully answer the question of what is the value of an extra roster spot.

10
by TheSportistician :: Tue, 03/22/2016 - 4:39pm

Agreed, but do beware of subtle recollection bias regarding these two players. We tend to remember the first and the last things we see/hear and forget the middle.

Sure, Kleinsasser didn't do much his first three years as a fullback after his 2nd round pick, but had a couple of good years (37-393-1 in 2002, 46-401-4 in 2003) once he started playing TE exclusively and ultimately had a 13 year career with his drafting organization.

And yes, Shiancoe's first year with the Vikings (278-323-1) was disappointing after his 5 year-18 million dollar free agent signing, but he did post an average receiving line of 52-525-6 in his last four years in Minnesota.

Sure, these aren't pro-bowl numbers, but it seems like their raw athleticism served them well in achieving better than average NFL careers (Career Approximate values of 14 and 23 respectively vs. the median 4 for combine participants between 1999-2011). This is what box-plots above are trying to illustrate: Pro-bowl athleticism may not always result in a pro-bowl selection, but it does appear to correlate with NFL success.

11
by Will Allen :: Tue, 03/22/2016 - 6:39pm

Shiancoe was so frustrating to watch, from the apsect of having great ability to gain seperation, but being really, really, mediocre at fnishing the play. He was one of those guys who didn't catch anything if he had any noise around him.

I think if Kleinsasser had not injured his knee in 2004 he almost certainly would have made a Pro Bowl or two or three. He was rounding into being a receiver with decent value, and was a simply devastating blocker.After the injury, his receiver vaue dropped a great deal, and he merely became a very, very, good blocker.

12
by TheSportistician :: Wed, 03/23/2016 - 6:02pm

Just ran the numbers for Hunter Henry using his Pro-day Dash and vertical: 2.1%. Though all his numbers are similar to Jason Witten, being 15 lbs lighter really hurt his final score.

13
by logan35 :: Fri, 07/01/2016 - 3:34pm