Effect of Weight gain on SAVVAGE score

After much deliberation, I've decided to re-brand my Size Adjusted Probowl Prediction score with a catchier acronym.  The metric detailed previously on this blog will be known from henceforth as Size Adjusted Velocity and Vertical Athletic Greatness Estimate, or SAVVAGE for short.  While I had some hope to add college production numbers to the metric, I've realized that the frequency of crossover athletes entering the NFL gives the simple version more value.  Please be aware this model is still being validated using prospective data analysis, but the model correctly predicted both Tyler Eifert and Travis Kelce in the validation phase of this metric already.

To explore the change in SAVVAGE due to weight gains, let's consider a couple of intriguing wide-receivers who, like just like Shannon Sharpe, converted to tight-end in their early NFL years.  Take these numbers with a grain of salt as applying this model to these players is an extrapolation since the training set was all NFL combine participants in the tight-end position from 2000-2011.

Darren Waller Combine Values (2015):
Vertical: 37 inches
Height: 78 inches
High Point:  115 inches
Dash:  4.46 seconds
Weight: 245 pounds
Avg Momentum: 54.9
Probowl odds: 1 in 4.3
SAVVAGE score: 18.7%

Darren Waller in 2016:
Weight: 260 lbs
Avg Momentum: 58.3
Probowl odds: 1 in 0.78
SAVVAGE score: 58%

Under the rather generous assumption that Waller maintained his ridiculous 37 inch vertical and 4.46 speed, this would be the 2nd highest known SAVVAGE score to date, being second only to Vernon Davis.   Despite the questionable assumptions, the change observed in his SAVVAGE score really illustrates the value of weight and speed for success as an NFL tight-end.  Still as it recently came to light that Waller tested positive for performance enhancing drugs, it's possible that most of this weight is muscle and he may have retained this ridiculous athleticism for the most part.

Let's also Niles Paul, another converted wide-receiver out of Nebraska, of slightly smaller stature and less ridiculous athleticism.  After gaining 26 pounds over his 5 years in the league, it's a bit absurd to think he's still as fast as he was at 224.  However, the goal is to explore how a hypothetical weight change can affect one's SAVVAGE score.

Niles Paul (2011 combine):
Vertical: 34.5 inches
Height: 73 inches
High Point:  107.5 inches
Dash:  4.51 seconds
Weight: 224 pounds
Avg Momentum: 49.7
Probowl odds: 1 in 479
SAVVAGE score: 0.2%

Niles Paul (2015):
Weight: 241 pounds
Avg Momentum: 53.4
Probowl odds: 1 in 64
SAVVAGE score: 1.5%

Niles Paul (2016):
Weight: 250 pounds
Avg Momentum: 55.4
Probowl odds: 1 in 22
SAVVAGE score: 4.3%

While Paul still doesn't quite top the 6% threshold that has historically done well at predicting Probowl appearances, it does go to show what a difference that weight plays in this size-speed equation.  The reader is encouraged to plug the combine metrics of an NFL tight-end into this calculator and play with the numbers a bit to see what changes correspond to big changes in their SAVVAGE score.

Bases Batted in 2015 MLB

As a follow-up to a previous post, I sat down to compute the number of bases generated by MLB player at bats as documented in the 2015 play-by-play data from Retrosheet.org.  Due to data limitations, the National League was neglected, though the error margins suggest that we're dealing with a large enough sample with the American League to serve as reasonable estimates for all MLB.

Here's are the aggregate stats detailing the frequency and probability of base-runner distribution, stratified by the base-runner positions:

2015 - AL	None	R@1st	R@2nd	R@3rd	R@1,2	R@1,3	R@2,3	R@1,2,3
Frequency	48498	15041	6234	2007	5403	2271	1509	1756
Probability	58.6%	18.2%	7.5%	2.4%	6.5%	2.7%	1.8%	2.1%
Margin of Error	0.4%	0.6%	0.7%	0.7%	0.7%	0.7%	0.7%	0.7%

Runner movements after each at-bat with runners on base were also tabulated, which are included merely for the purpose of making these results reproducible/verifiable:

FROM 1ST	R@1st	R@2nd	R@1,3	R@1,2,3
Runner lost	2560	765	372	232
No Advance	7623	2993	1142	893
Batted to 2nd	2704	924	421	367
Batted to 3rd	1269	392	199	157
Batted Home	835	312	122	94
Hm - unearned	50	17	13	13
Hm - unearned team	0	0	2	0
FROM 2ND	R@2nd	R@1,2	R@2,3	R@1,2,3
Retired Side	134	176	22	34
Runner lost	3843	3182	976	1016
Batted to 3rd	1184	1011	243	327
Batted Home	992	944	238	341
Hm - unearned	79	89	29	36
Hm - unearned team	2	1	1	2
FROM 3RD	R@3rd	R@1,3	R@2,3	R@1,2,3
Retired Side	36	56	39	79
Runner lost	1307	1383	938	1015
Batted Home	608	737	474	591
Hm - unearned	56	94	57	71
Hm - unearned team	0	1	1	0

After excluding unearned advancement home, all remaining base advancements were given a positive weight of 1, with the appropriate negative weight applied for runners lost from 1st, 2nd, and 3rd respectively (-1, -2, -3) due to a force-out, tag-out, or when the side was retired.  In this way, the total bases gained/lost after each of these at bats was tabulated under each runner condition:

2015 - AL	R@1st	R@2nd	R@3rd	R@1,2	R@1,3	R@2,3	R@1,2,3
Frequency	15041	6234	2007	5403	2271	1509	1756
Total Bases	5187	2900	500	4426	1382	1032	2026
E(bases)	0.34	0.47	0.25	0.82	0.61	0.68	1.15
Margin of Error	0.017	0.021	0.028	0.018	0.026	0.028	0.025

Using these averages as approximations for the expected bases from each condition, we can might compare the results of a batted bases of an individual player to that of the league average, but that seems like a matter of another post.

New Touchback Rule May Increase Kick-Returns

While the kick return is perhaps the most exciting play in football, it is also one of the most dangerous.  You don't have to be Sir Isaac Newton to know that players hurling themselves at each other while running full-speed in opposite directions is an accident waiting to happen.  The particularly heinous detail here is that special team players are often the very minimum salaried players that the NFL should work the hardest to protect.

As such, hoping to increase touchbacks and decrease injury, the NFL moved the kickoff location forward from the 30 to the 35 yard line in 2011. Indeed, as illustrated in the associated figure, this resulted in a dramatic decrease in the relative percentage of kicks returned.  As the margin of error bars illustrate, the percentage of kicks returned has been steadily declining ever since 2011, with even the 2015 season appearing statistically significantly lower than that of 2014. Note that the data was derived from aggregate kickoff info obtained on www.footballdb.com.

However, there are questions as to if this improvement was enough.  This figure was generated in an attempt to visualize all kick returns from the 2015 season, which is sorted by location that the kick was originally fielded. This data was extracted from downloaded play-by-play information complied by nflsavant.com, which has the added benefit of allowing us to extract when an injury occurred on particular kick return or touchback.  Note that cases of falling on an onside kick and lateral situations were excluded due to the respective simplicity and complexity, but pose perhaps greater risk. 

In fact, in the three years of play-by-play data available on nflsavant.com, there were still 53 injuries on 2974 kicks (1.75% injury rate) vs. 13 injuries on 3297 touchbacks (0.33% injury rate).  These respective percentages imply that the risk of an injury is 5 times greater on a return than on a touchback.  And as the above figure indicates, more than 50% of returns are coming out of the end-zone, so if we could give returners greater incentive to down the ball, we might expect the number of injuries to decrease by around 50%.

Maybe cutting 53 injuries by a factor of 2 doesn't sound like a great saving to the average fan, but let's remember how terrible each of these injuries can be.  In a week 4 kick return last season, the Cowboy's Lance Dunbar tore his ACL/MCL/patellar tendons on a return to begin the second half. It is important that we remember that there is an individual behind each of these statistics that is suffering a life-altering and potentially career ending injury.

In an effort to further protect players, the NFL announced this off-season that they would be modifying the location of touchback to the 25 yard-line after a kickoff in the 2016 season.  The hope is that this probationary rule provides an additional 5 yard incentive for returners to take a lower risk touchback.  If we increase touchbacks, it will theoretically reduce injuries in special teams players.

In fact, if we stratify the returns from the past three years based on those started from the end-zone and those from outside, we see some very different distributions in play. Clearly returners have incentive to down a kickoff caught in the end zone.  However, with the median of the "out of EZ" distribution lying at 25 yards exactly, it's NOT clear that kickers would have incentive to keep booming their kicks to the end zone and beyond.

Still, not all kicks in the above distributions are created equal.  The difference between deep kicks and shallow ones should have a dramatic effect on the resulting starting field position.  If we further stratify our data by position the kick was fielded, we can bin how the return fared relative to the new 25 yard touchback location.  Again, the error bars are 95% margins of error that allow us to tell if, in spite of the variation due to randomness, the average starting field position after the return is at or beyond the 25 yard line more than 50% of the time.

This breakdown suggests that the average kicker may minimize his opponents starting field position by attempting to kick exclusively between the 0 and 5 yard line.  Still, as with all observational data, we may be dealing with a bias of NFL kickers only attempting to pin it in the 0-5 range with less than average returners.  If this is attempted with more skilled returners in the league, they may find this strategy is actually less than optimal.  In the end, some Bayesian methods based on prior return yardage may help kickers decide in real time which returners will get a booming kick and which a mortar.  However, at the onset, many returners are unknown NFL commodities, so the prior of league average may suffice.  Thankfully the league is using the next year to evaluate the rule in a probationary period so they may be able to easily abandon the rule if it has the adverse affect than intended.  Still, this may be after we have already put some special players at additional risk.