Adjusted SPARQ: Position-Adjusted Athleticism

Why “Adjusted SPARQ” is Needed

SPARQ is an athletic score originally built by Nike that’s created by using speed, strength, agility, and size data points. It’s easy to translate SPARQ scores from NFL Combine data, which has made the athletic composite score very popular in recent years. SPARQ scores are helpful for a prospect’s evaluation, but it’s a flawed metric.

SPARQ was designed to measure overall athleticism, not specific athletic traits. That’s totally fine for some sports, but the NFL is far too specialized to have one single formula determine how athletic a player is. For example, the SPARQ formula doesn’t change when measuring a defensive tackle and cornerback, but we can all agree that the “athleticism” required for each position is totally different. A DT needs more strength and size, while a CB needs more speed and agility. These position adjustments are necessary to measure true athleticism for an NFL prospect, and that’s why I created “Adjusted SPARQ”.

“Adjusted SPARQ” Methodology

I built multiple linear regression models for each NFL position that predict how much production a prospect will have during his first four NFL seasons (a.k.a. rookie contract production). NFL Combine and Pro Day measurements were my regressors and Pro Football Reference’s Approximate Value (AV) metric was my output variable. I limited the dataset to players who were drafted from 2005-2016.

The coefficients of the input variables change from position to position because the value of each athletic test differs based on what the position is asked to do. I’ve included the “Adjusted SPARQ” formulas and r-squareds below. Also note how I split some positions (RB, WR, DT, EDGE) into different groups based on size. I did this because Hunter Renfrow and DK Metcalf shouldn’t be viewed as playing the same position. They are asked to do two totally different things, so it’s best to build separate models for the two.

“Adjusted SPARQ” Formulas

The correlations of Adjusted SPARQ to rookie contract success vary from position to position, but these scores have higher correlations than traditional SPARQ so it’s worth transitioning to these formulas in my opinion. Still, it’s important not to overrate adjusted athleticism because it only explains 5-25% of a player’s success in the NFL.

RB (under 210 pounds)

Adjusted SPARQ	Coefficient	P-Value
Constant	39.2933	0.52
BMI	3.64339	0.00
10 Yard Split	−30.6156	0.31
Three Cone	−12.5660	0.12

Adjusted R-Squared

SPARQ = 0.00 (aka completely useless)

Adjusted SPARQ = 0.15

Conclusion

Traditional SPARQ is completely useless when it comes to predicting NFL success for undersized NFL running backs, but my Adjusted SPARQ formula found that Body Mass Index (BMI) is a little important, however. Specifically, it’s looking for thicc prospects who have above-average agility (Three Cone) and short-area burst (10 Yard Split). The classic example of an undersized running back who checked these boxes was Ray Rice.

RB (at least 210 pounds)

Adjusted SPARQ	Coefficient	P-Value
Constant	31.3032	0.61
BMI	−1.94016	0.02
Cone	−2.39364	0.69
Speed Score	0.24573	0.07
Broad	0.273107	0.29

Adjusted R-Squared

SPARQ = 0.03

Adjusted SPARQ = 0.13

Conclusion

Traditional SPARQ doesn’t do a good job of predicting NFL success for bigger running backs, but my Adjusted SPARQ formula found that weight-adjusted speed (Speed Score) and explosion (Broad) were the highest correlated athletic traits for big backs, while agility came in as a minimal factor. The model does ding prospects who carry excess weight as well (BMI). Two prospects who checked all of these boxes were Matt Forte and Adrian Peterson.

WR (under 6’0)

Adjusted SPARQ	Coefficient	P-Value
Constant	67.4649	0.30
Speed Score	0.266259	0.08
10-Yard Split	−27.2128	0.31
Short Shuttle	−9.34978	0.28

Adjusted R-Squared

SPARQ = 0.00 (completely useless)

Adjusted SPARQ = 0.07

Conclusion

Athleticism isn’t very important for undersized receivers. Of all the metrics my model looked at, the weight-adjusted forty (Speed Score) carried the most weight, while agility (Short Shuttle) and short-area burst (10 Yard Split) were minimal contributors. Ultimately, it’s not the end of the world if a sub-six foot receiver doesn’t shred the NFL Combine. Two prototype sub-six foot receivers are Brandin Cooks and John Brown.

WR (at least 6’0)

Adjusted SPARQ	Coefficient	P-Value
Constant	-146.167	0.01
Height	1.12291	0.10
BMI	0.93771	0.19
Broad	0.401825	0.01

Adjusted R-Squared

SPARQ = 0.04

Adjusted SPARQ = 0.05

Conclusion

Putting too much weight into NFL Combine scores for big receivers can be problematic. Both traditional SPARQ and Adjusted SPARQ do a fairly poor job of predicting NFL success, but my model did find burst (Broad Jump) and size (Height, BMI) to be a little helpful. Putting any weight into agility scores for tall receivers is a big mistake with the most recent example being DK Metcalf. Players like Dez Bryant and, of course, Julio Jones checked these boxes as big receiver prototypes.

TE

Adjusted SPARQ	Coefficient	P-Value
Constant	−39.8474	0.41
Height	0.779761	0.16
Vertical	0.201368	0.48
Cone	−4.75801	0.19
Speed Score	0.137361	0.09

Adjusted R-Squared

SPARQ = 0.05

Adjusted SPARQ = 0.08

Conclusion

Among the athletic measurables for tight ends, there’s one key thing to pay attention to: the weight-adjusted forty (Speed Score). That metric carries the most significance, while size (Height), agility (Cone), and burst (Vertical) were secondary components. If a tight end prospect can’t run, it’s not looking good for his NFL future as a pass-catcher. Jared Cook and Travis Kelce are examples of tight end prototypes, at least if you’re looking for a producer and not a hand in the dirt blocker.

OT

Adjusted SPARQ	Coefficient	P-Value
Constant	-25.0741	0.70
Arm Length	0.725946	0.50
Weight	0.349363	0.03
BMI	-1.36164	0.23
Forty	-13.9585	0.06
Bench	0.218972	0.35
Broad	0.223903	0.30

Adjusted R-Squared

SPARQ = 0.05

Adjusted SPARQ = 0.08

Conclusion

The offensive tackle position is complicated because the best players have a mixture of physicality (Weight, Bench) and speed (Forty). Being lengthy and not overly heavy relative to a tackle’s height are important factors, too. My model surprisingly excluded any agility measurables, which was a surprise to me given their pass-blocking technique, but being big and fast has proven to be more important over the last decade-plus. Offensive tackle prototypes are Nate Solder and Jack Conklin.

iOL

Adjusted SPARQ	Coefficient	P-Value
Constant	18.4942	0.51
Shuttle	-8.34605	0.06
Speed Score	0.171192	0.06
BMI	0.518003	0.33

Adjusted R-Squared

SPARQ = 0.02

Adjusted SPARQ = 0.03

Conclusion

From an analytics perspective, it’s almost impossible to predict NFL success for interior offensive linemen. Production metrics are hard to come by and athleticism isn’t a good predictor of rookie contract production. Both traditional SPARQ and my Adjusted SPARQ have proven to be bad, but the weight-adjusted forty (Speed Score) and agility (Shuttle) would be the things to pay attention to if forced to look into interior line athleticism metrics. I guess Nick Mangold and Ryan Kelly are prototypes.

DT (under 310 pounds)

Adjusted SPARQ	Coefficient	P-Value
Constant	-4784.00	0.00
Height	49.3106	0.01
Weight	-7.28111	0.01
BMI	48.8388	0.02
Forty	235.314	0.03
Shuttle	17.5122	0.10
Cone	-17.9686	0.01
Speed Score	3.40701	0.01

Adjusted R-Squared

SPARQ = 0.07

Adjusted SPARQ = 0.20

Conclusion

My Adjusted SPARQ model has a notably higher correlation to NFL success for “undersized” defensive tackles than traditional SPARQ. The model uses a combination of size, speed, and agility while ignoring the bench press and jumping tests. Defensive tackles who are quick and agile are the ones who test the best in my model. Aaron Donald and Fletcher Cox are prototypes.

DT (at least 310 pounds)

Adjusted SPARQ	Coefficient	P-Value
Constant	-1245.76	0.00
Arm Length	3.07869	0.02
Weight	-1.44485	0.00
BMI	2.38214	0.04
Forty	231.269	0.01
Speed Score	3.66397	0.00

Adjusted R-Squared

SPARQ = 0.01

Adjusted SPARQ = 0.20

Conclusion

Athleticism definitely matters for defensive tackles. Similarly to their undersized teammates, big defensive tackles require a lot of size and speed. That’s why BMI, arm length, and the weight-adjusted forty are accounted for in my model. Agility and jumping scores, meanwhile, are completely ignored, as they hold nearly zero predictive power. Thicc defensive tackle prototypes include Dontari Poe and Linval Joseph.

EDGE (under 270 pounds)

Adjusted SPARQ	Coefficient	P-Value
Constant	26.5137	0.47
Broad	0.468644	0.01
Cone	-12.3589	0.00
Speed Score	0.177537	0.08

Adjusted R-Squared

SPARQ = 0.13

Adjusted SPARQ = 0.18

Conclusion

Athleticism clearly matters for edge rushers. Explosiveness (Broad, Speed Score) and agility (Three Cone) specifically are key for speed rushers who need to bend around offensive tackles or counter inside to create pressure on the quarterback. I did find it interesting that a size or strength component wasn’t factored into my model, suggesting it’s okay to draft potentially undersized edge rushers as long as their mobile. Von Miller and Danielle Hunter are prime examples of this type of player.

EDGE (at least 270 pounds)

Adjusted SPARQ	Coefficient	P-Value
Constant	1531.33	0.03
Weight	1.79504	0.01
Forty	-305.213	0.03
Twenty Split	-42.4833	0.22
Vertical	1.0816	0.03
Cone	-13.5584	0.04
Speed Score	-3.48922	0.03

Adjusted R-Squared

SPARQ = 0.20

Adjusted SPARQ = 0.27

Conclusion

Finding oversized edge rushers using athleticism is a little more complex than finding speed rushers, but this model does a good job of finding intriguing prospects. Size and speed are the two major components of the formula, while explosiveness (Vertical) and agility (Cone) play secondary roles. Getting that all-around stud athlete is the goal with this position because it requires multiple ways to win (bull rush, speed rush, etc.). Mario Williams and J.J. Watt are obvious prototypes.

LB

Adjusted SPARQ	Coefficient	P-Value
Constant	65.4583	0.18
10 Yard Split	-23.5360	0.24
Shuttle	-9.59481	0.12
Speed Score	0.270717	0.01

Adjusted R-Squared

SPARQ = 0.05

Adjusted SPARQ = 0.10

Conclusion

Most of the predictiveness in athleticism for off-ball linebackers can be explained by the weight-adjusted forty (Speed Score). It’s by far the most helpful athletic metric I’ve come across, one that has pinpointed the best linebackers of the 2000s. My model also incorporates agility (Shuttle) and burst (10 Yard Split), but sorting by speed score is nearly as effective. Linebackers like Patrick Willis and Bobby Wagner are prototypes for the position. I won’t be surprised if linebackers continue to become smaller and faster as passing games and sweeps become more popular in NFL offenses.

Safety

Adjusted SPARQ	Coefficient	P-Value
Constant	783.437	0.03
Weight	0.992385	0.02
Forty	-163.288	0.04
10 Yard Split	-26.5352	0.12
Vertical	0.435219	0.12
Shuttle	-6.57705	0.20
Speed Score	-1.87349	0.04

Adjusted R-Squared

SPARQ = 0.03

Adjusted SPARQ = 0.05

Conclusion

Safety is another position that’s difficult to evaluate because college production and athleticism aren’t very predictive of NFL success. Size, speed, agility, and burst all play minor roles in my Adjusted SPARQ model, but none of the available metrics are that important. If forced to pick metrics to look at, I’ll focus on raw speed (Forty, 10 Yard Split) and burst (Vertical). Eric Berry and Eric Reid are safety prototypes.

CB

Adjusted SPARQ	Coefficient	P-Value
Constant	-27.2002	0.67
Height	1.45033	0.01
BMI	2.01554	0.00
10 Yard Split	-24.3512	0.12
Vertical	0.348413	0.17
Cone	-4.18381	0.23
Forty	-14.6466	0.14

Adjusted R-Squared

SPARQ = 0.07

Adjusted SPARQ = 0.09

Conclusion

Athleticism matters a little more for corners than it does at safety, but relying on athleticism is a tad problematic. My Adjusted SPARQ model found size (Height, BMI) and speed (Forty, 10 Yard Split) to be the most important factors, while agility (Cone) and burst (Vertical) played minor roles. Jalen Ramsey and Byron Jones are examples of big and fast corners that my model loves.