More Kick and Punt Return TDs are Surrendered Early in the NFL Season

I was recently reading about football as I often do. For the uninformed reader, my preference is for the collegiate game but I do possess a certain fondness for the professional game. It fascinates me, the extent to which many professionals prepare for and intellectually categorize the intricacies of such a violent game. The machinations of pro football that occur away from the gridiron, on days other than Sunday and during the offseason are intriguing as well. This, notwithstanding the machinations of pro pigskin that I detest.

Anyhow, I was reading Take Your Eyes off the Ball authored by Pat Kerwin. I enjoyed his anatomization of team management activities in the League. In brief, it is a quality read for the novice football fan interested in learning about Xs and Os, team management operations, or both.

In his discussion of special teams play, Kerwin asserts that we “often see a flurry of punts and kickoffs returned for touchdowns [during] the first three weeks of the season.” He attributes this effect to poor management of practice time, roster turnover and the limitations of a 53-man roster, and coverage teams often being comprised of inexperienced players.

I was interested in testing Kerwin’s assertion. Data were culled from the Football Reference Play Finder for seasons 2009-15. I divided the 16-game NFL regular season into quadrants: team-games 1-4, 5-8, and so on, and computed six variables for each quadrant: 

• punt return (PR) and kickoff return (KR) TDs; 
• PRs and KRs that were not TDs; 
• and punts and kickoffs with no return (NR).

Table 1. Observed and Expected Values for NFL Kickoff Outcomes 2009-15

		OBSERVED			EXPECTED
GAMES	Kickoffs	KRTD	KR	NR	KRTD	KR	NR
1-4	4406	18	2171	2217	16.8	2489.4	1899.8
5-8	4467	21	2480	1966	17.0	2523.9	1926.1
9-12	4382	14	2620	1748	16.7	2475.9	1889.5
13-16	4343	14	2672	1657	16.5	2453.8	1872.6
TOTAL	17598	67	9943	7588

Data for KRs and PRs appear in Table 1 and 2, respectively. The values included are the quantities of TDs, returns, and NRs that were observed and those that would be expected given the proportions that emerged. This arrangement is suited for a Chi-square test of independence because that is the analysis we will use. Though excluded for visibility, do note that expected-column totals equal that of observed-columns.

Table 2. Observed and Expected Values for NFL Punt Outcomes 2009-15

		OBSERVED			EXPECTED
GAMES	Punts	PRTD	PR	NR	PRTD	PR	NR
1-4	4137	30	1833	2274	24.3	1807.9	2304.8
5-8	4193	19	1874	2300	24.6	1832.4	2336.0
9-12	4295	27	1892	2376	25.2	1876.9	2392.8
13-16	4233	23	1768	2442	24.9	1849.8	2358.3
TOTAL	16858	99	7367	9392

KR TDs are observed more than would be expected during the 1-4 and 5-8 team-games of the season and less than would be expected in the latter quadrants, χ²(6, N = 17,598) = 160.34, p <.001, φ_C = .067. Given the effect size (φ_C), we conclude that there is a small but significant disparity in the distribution of KR TDs throughout the season. Although we observed differences in PR TD distributions, the distributional disparities through the season appear minimal, χ²(6, N = 16,858) = 11.98, p = .062, φ_C = .019. However, it is worth noting that PR TDs during the first four games of the season are probably occurring more than would be expected, supporting Kerwin’s hypothesis.

Chi-square tests indicate that Kerwin is accurate in his assertions. NFL teams surrender more KR and PR TDs early in the season. However, this analysis merely tests the distribution of observed PR/KR TDs and what would be expected to happen given those observations. That is, we are unable to test the influence of poor management of practice time, roster turnover, inexperience, etc. The present analysis does provide a foundation for future studies of the foregoing variables.

Comparison of WNBA and NBA League-wide Team Data

Monday 30 MAY 2016. Update: I prepared and published this post in a haste only to realize while in the shower several days later the mistaken inclusion of three variables in the Table of this post: "ORB / FG missed", "DRB / FG missed", and "Dead Ball Rebounds". With the data that was used in computation, these are inaccurate and meaningless statistics because rebounds can be gathered on missed shots as well as missed freethrows and we are unable to distinguish from league and team totals whether rebounds were gathered on FGAs or FTAs. Likewise, Dead Ball Rebounds was inaccurately included in my haste and computation of such a statistic would require additional data. Apologies.

Point guard of the WNBA Seattle Storm Sue Bird contributed to a recent installment of the 538 podcast HotTakedown. Bird’s appearance was precipitated by an op-ed she authored for the Player’s Tribune illuminating the paucity of accessible, informative data for the WNBA and for female sports more broadly. Days later, hysteria followed comments made by a veteran sports writer regarding the unprecedented dominance of the UConn women basketball team. The implications of that writer’s statement have been summarized elsewhere.

After listening to Sue Bird on the 538 podcast and before learning of the galvanizing comments, I watched much of Uconn’s decimation of Mississippi State. I watched because I enjoy that sort of competitive dominance; plus, their team is just good. I’ve yet to write about it here but I also watched because I love basketball. After hearing the galvanizing comments, all I could think was: with increased access to informative data and with an array of perspectives creating narratives using such data, maybe a weakness would be identified.

Admittedly, I watch less female than male basketball but that is partly attributable to the greater viewing options for the latter. In my opinion, the female game demands greater acceptance of and adherence to strategy. Also, it appears that fewer female players exhibit an inclination to rely on and expect officiating. There are absolutely aspects of female basketball that I prefer. That should be unsurprising—consider the blog title—because there is simply less palming of the ball, at least in my observational comparisons of the female and male varieties.
 
Indeed, there is diversity and variety in the female and the male games. This post is devoted to a simple exploration of that variety. The data in Table 1 was computed using league-wide WNBA and NBA data from the 2015 and 2014-15 regular seasons, respectively. Data from Basketball Reference. Possessions were estimated with an equation used by ESPN.com and NBA.com, developed I believe by legend Dean Oliver. 

Table 1. Comparison of WNBA and NBA League-wide Team Data for Regular Season Ending 2015

SHOOTING
STAT	WNBA	NBA
FG%	42.5%	44.9%
3FG%	32.5%	35.0%
2FG%	45.4%	48.5%
FT%	79.5%	75.0%
% of FGA are 2PA	77.4%	73.2%
% of FGA are 3PA	22.6%	26.8%
ORB / FG Missed	23.4%	23.6%
DRB / FG Missed	65.9%	70.4%
Dead Ball RBs	10.7%	6.0%
Assists / FGs	58.7%	58.7%
Blocks / FGA	6.6%	5.7%
FT / FGA	22.6%	20.1%
FTA / FGA	28.4%	27.3%
POSSESSIONS
FGA / POSS	0.877	0.897
Fouls / POSS	0.248	0.217
Turnovers / POSS	0.173	0.154
Steals / POSS	0.096	0.083
Pts / POSS	1.008	1.073
Pts / FGA	0.923	0.992
Pace	88.7	92.5

A few items of note upon review of this table. Offenses in both leagues rebound their own misses at similar rates and assist FGs at essentially identical rates. The pace of the NBA game is somewhat faster. In the WNBA there appears to be a greater proportion of dead-ball rebounds the cause of which is unclear. Also, for what it’s worth, the NBA teams scored .069 points per possession more than did WNBA teams. 

Before concluding, I would like to provide some depth to the discrepancies in FG% between the two leagues. NBA players tallied 8793 successful dunks in 2014-15 season. From what I can ascertain, there were maybe 2 dunks in the 2015 WNBA season. Successful dunking is nearly guaranteed, although NBA did make only 91% of dunk attempts. Contrarily, NBA players sunk 27,080 lay-ups that season at a clip of 58%. As you know, we don’t have that sort of data for the WNBA; well, at least I don’t. 

So how does the discrepancy change if we exclude all dunk attempts from the computation of NBA FG%? We’ll even remove the 2 WNBA dunks. Excluding dunk attempts, NBA FG% drops from 45% to 40%. Considering that FG%, WNBA players appear to shoot a higher percentage that is statistically significant, but it is a small effect.[1] Now, it should be noted that the WNBA players do play with a slightly smaller and lighter-weight ball, about 96.6% the circumference of the NBA ball and 91% the weight of the NBA ball but the rim diameters are equal. Thus, the argument could be made that there is a greater area of the basket plane available for the WNBA ball to enter the basket. 

There is evidence to suggest that basketball of light weightwill facilitate increased FG%. We do know that Illinois males played Oakland in December 2010. The first 7:22 of game time was played with the slightly smaller ball used by females. Illinois did shoot 3 of 13 in that time and Oakland shot 7 of 16. That is a small sample, however, there appears to be a lack of effect of ball size on free throw shooting percentage and on shooting kinematics. Moreover, one study compared collegiate 71 female and 35 male basketballers. Neither an effect of ball size nor weight was found for on “side shots” (elbow jumper) and “lay ups” but smaller, lighter balls were passed fastest by both sexes. 

I provided in this post a comparison of the WNBA and NBA league-wide team data. However, I was not attempting a comparison, per se. More precisely, I attempted to highlight the variety of game-play in each League and how that variety manifests statistically because, when informative statistics are tracked for female basketball, it is not enough to simply adopt existing metrics from the male game—ecological considerations specific to the female game are warranted. Simple calculations were used to demonstrate how the high-percentage of successful dunking inflates FG%s in the NBA. Likewise, the effect of ball-size was considered. In sum, this exploratory analysis indicates several avenues of further study of the statistical manifestations of female basketball including, the high proportion of dead ball rebounds and slight differences in turnovers, fouls, steals, and blocked shot-attempts to their male counterparts.

---

[1] χ²(df = 1, N = 27055) = 103.35, p < .001, Ṽ  = .061.

Personal Anecdote about Megatron

His likeness appears in the logo of this blog. It is only right.

Calvin Johnson announced today his retirement from NFL. His was an illustrious career on the field—a career fraught by disappointing team outcomes; a career ended gradually by physical abuses incurred fighting double- and triple-coverage. Johnson was the sole purpose for my tuning into a Detroit Lions game at any point in my adult life.

Prior to my adult life was an adolescent phase of impetuosity and intemperance. Conjoining these two eras of my life was a period of reformation during which I focused on slam dunking. At age 12 I was 5’1” and could tap a regulation backboard; my father, 6’, last dunked at age 35, with a volleyball. I loved basketball and I loved the dunk. I remember doing 360s on those FisherPrice hoops. So it was natural at that point when I needed it most. At present I am slowly crafting a cohesive account of the slam dunk including its history, its suitability as an Olympic event, and standardized contest protocols and rank assignations.

Anyhow, circa 2006 I was 20 or 21 and immersed in cultivating my ability to dunk. Hint: it was never spectacular, especially after losing my abs and glutes to grad school. Near to my childhood home was a basketball court that was often glazed with sand. The court opened to an inland waterway dotted with commercial and recreational watercrafts. Surrounding the court was a playground as well as a rickety building and facilities fit for parties and cookouts. There was a water fountain that dispensed warm, brackish water. Since 1979, my hometown and surrounding area has hosted one of many youth sport camps replete with former professional players and coaches. Unbeknownst to me, in 2006, several players from the Georgia Tech football team were involved in the camp. 

Calvin Johnson dunked on that hoop in 2006.

It was an average June day in the south. It was fucking hot. And probably really humid. I went to the aforestated court and noted a teeming crowd upon my arrival. My warm up probably did not deviate from my usual routine of freethrows, end-to-end lay-ups, and HORSE shots.

After I started dunking—y’know, some simple two-handers and one-handed lobs—a man about my height but noticeably bulkier beckoned for the ball. He and I took some shots amidst minimal conversation. A group of guys accompanying him sauntered onto the court. I will add that I was disengaged in college football at this point and really, any sport other than dunking basketballs.

When that man went up for his first dunk, the ball ricocheted off of the backrim into the blue sky and the rim uttered a shriek before locking in the down-position, which it often did. The goal shook for what seemed like an actual minute. Some of the other guys and myself dunked, in sort of a cipher. An audience of kids and parental units congregated about the court. They applauded our performances. It was kind of exciting for me. Some of the guys complimented me on my jumping ability before the crowd dissipated and the bunch departed to devour hamburgers and the rest. 

Tashard Choice.

I later learned that then running back at Georgia Tech Tashard Choice was the man my height. I thought it was neat to have dunked with him since he was a stand-out in college and played professionally for some time but I never revisited the experience until a few years ago. Throughout our exhibition there was a tall, reserved member of the group who was utterly jacked. All of them were, but he was remarkably chiseled. I realized at that point some years later that the quiet one, I am almost certain, was Calvin Johnson. He dunked twice, I believe. Before the audience came, he corralled someone’s rebound and threw in a simple one-handed dunk. I recall him commenting concisely on the rim height being less than regulation, an observation that I confirmed; it was about 9’8”. After the crowd assembled, Choice commanded everyone’s attention due to the sheer power with which he dunked. Calvin Johnson’s second dunk that day came in front of the crowd. It was a windmill following a short walk-up and I believe it was one-handed. Everyone cheered. He is 6’5” and the rim was <10’ but it was soooo effortless for him, and his strength was evident in the trembling hoop.

In summation, that was a ramble about unknowingly encountering a unique physical talent, Calvin Johnson, among a group of unique physical talents, division 1 football players—at least two of which played professionally. Standing 6’5” and weighing ~240lbs, Johnson was even more confounding and perplexing offensive matchup due to his 4.35 second forty speed. And it is unsurprising his effortless dunking that day nearly 10 years ago considering his 42.5” vertical leap.  Football is a game, but it is game I love and it is unfortunate that he is unable to remain young forever.

Field Position Part 1: Interception Return Yards

In a previous post I mentioned my intent to explore the effects of defense and special teams on field position. This is part one of that investigation. In another post I offered a method of valuating the average yardage of an interception (for the intercepting team). At present, I will discuss a different method of calculating and valuating INT yards using play-by-play (PBP) data.

Yes, collegiate and professional statistical records include yardage gained on interception returns just as passing or rushing yards are included. However, we track passing and rushing yards as a measure of progress and productivity—a measure of player or team performance. We track interceptions because each signifies an exchange of possession. The interception itself is the measure of player or team capacity—not the yardage gained on returns. Interception return yards are unexpected gratuity.

Chris Harris picks off and returns a Kyle Orton pass.

For instance, consider two interceptions from 2014. A Tony Romo pass was intercepted by NY Giants’ Prince Amukamara and Buffalo Bill Kyle Orton’s pass was intercepted by Dever Bronco Chris Harris. Amukamara and Harris were each credited with 38 INT return yards. Amukamara’s half way through the second quarter of a tie game and Harris’ with 5 minutes remaining in the third quarter, his team leading 21-3. Both players’ offenses scored on their following drives. So what differentiates Amukamara’s INT return from that of Harris? Field position.

Amukamara intercepted Romo’s pass at the NYG 35-yard line and returned the ball to the Dallas 27. Harris intercepted Orton’s pass at the Bronco 2-yard line and returned it to the Bronco 40. Indeed, Harris’ INT may be more valuable because he ended the possession of an opponent in scoring position.[1] But, Amukamara’s 38 INT return yards put his offensive teammates in field goal position before they lined up.

An offense gaining no more than 2.3 yards on every play of every game would be disbanded. A DB intercepting one pass per game and being downed at the spot of each INT would receive a max contract and an eponymous island. No writers would denigrate him for failing to gain yardage after the INT. Any coach or fan would prefer an INT returned to the 50 than one returned to his own 3, of course; but surely every coach or fan would prefer an INT to the opponent having possession. 

To me this means that, although there is value in knowing the yardage gained from the spot of an interception to the spot an interceptor is downed, ultimately, it is more meaningful knowing the field position produced by that gained yardage. That is, both players in our example should be credited 38 INT return yards but the values in the game at the point of each player being downed were more accurately described as 73 or 40. This is particularly true if we desire statistics that reflect happenings on the field.

I ran a Pro-Football Reference Game Play search for all interceptions in 2014 regular season excluding pick-sixes and interceptions with lost fumbles on the return. Pick-sixes were excluded because a touchdown precludes an offense driving and, thus, are uninvolved in the tabulation of average starting field position. Four returns with fumbles lost by the interceptor and recovered by the intercepted team were excluded because possession was regained.

Extracted from that data were the (a) spot of the interception and (b) spot of being downed following INT return. Computed with those values were the (c) interception yards from the spot of the INT to the spot of being downed or 20 for touchbacks and (d) starting field position for the interceptor’s offensive unit measured from a team’s goal line to b, the spot of being down. All spot-yardage values were scaled from 1 to 99 with 1 being the intercepting team’s own goal line and 99 being their opponents’ goal lines.

Table 1 contains various interception statistics for 2014 NFL teams, including League averages. The interception return yardage value we are interested in is Mean INT FP column. Teams are ranked by average starting field position following interceptions. Interestingly, I was forced to revisit my earlier debate of the greater value of Amukamara’s and Harris’ interceptions. It appears that over the course of the 2014-‘15 seasons, the Giants’ defense endowed their offense with the greatest field position advantage with interceptions but the average spot of their 16 interceptions was nearly midfield. Compare this to the average spot of the 16 Dallas Cowboys interceptions, their own 30—where opponents are within field goal range. Although intuitive, interception spot increased with field position following interception (N = 393, r = .81, p < .001). Interception return yards also increased with field position following interception (r = .41, p <.001).

Table 1. Interception Yards and Interception Field Position (Yards) Following Interceptions for 2014-15 NFL Teams

TEAM	Mean FP	Non-TD INT	Mean INT Spot	Non-TD INT Yards	Mean nTD INT Yards	INT FP Yards	Mean INT FP
SDG	26.4	6	18.8	72	12.0	185	30.8
CAR	27.7	10	23.4	190	19.0	424	42.4
NYJ	27.8	6	24.7	54	9.0	202	33.7
KAN	29.3	5	27.0	76	15.2	211	42.2
CHI	25.9	13	30.1	142	10.9	533	41.0
PIT	25.7	7	30.1	127	18.1	338	48.3
DAL	28.9	16	30.3	149	9.3	634	39.6
NWE	25.5	12	30.9	140	11.7	511	42.6
STL	28.0	10	31.3	108	10.8	421	42.1
ATL	26.5	15	32.1	104	6.9	586	39.1
MIA	31.1	11	33.1	188	17.1	552	50.2
SEA	27.8	21	33.5	298	14.2	1001	47.7
IND	28.7	11	34.5	96	8.7	475	43.2
SFO	30.5	11	34.6	207	18.8	588	53.5
PHI	30.0	9	35.3	56	6.2	374	41.6
JAX	25.7	5	35.8	65	13.0	244	48.8
CLE	26.8	18	37.9	260	14.4	943	52.4
ARI	26.9	15	38.3	154	10.3	728	48.5
BAL	28.7	10	39.1	91	9.1	482	48.2
DET	29.9	18	39.3	336	18.7	1043	57.9
NOR	30.6	16	39.3	200	12.5	829	51.8
GNB	28.5	15	39.5	228	15.2	820	54.7
CIN	30.3	19	40.4	174	9.2	941	49.5
WAS	25.1	6	40.5	30	5.0	273	45.5
MIN	27.3	11	40.8	90	8.2	539	49.0
BUF	30.2	18	41.2	346	19.2	1088	60.4
TEN	25.8	11	41.7	119	10.8	578	52.5
TAM	26.6	11	41.9	76	6.9	537	48.8
DEN	28.9	16	42.1	173	10.8	846	52.9
HOU	27.7	16	44.0	265	16.6	969	60.6
NYG	28.2	16	45.5	266	16.6	994	62.1
OAK	24.2	9	53.4	76	8.4	557	61.9
LEAGUE	27.9	12.3	36.8	154.9	12.6	607.7	49.5

Note: FP = Field Position. NON-TD INT = the yards produced on all team interceptions that do no result in TDs; this would be the values reported in League statistics minus yardage from pick-sixes.

---

[1] Pro-Football Reference’ Expected Points model tells us that Amukamara’s INT was worth -3.78 EPA and Harris’ -1.6 EPA but Amukamara’s INT yielded a net EP -3.56 and Harris’, -5.91. Amukamara’s INT is worth a greater EPA value probably due to resultant field position but Harris’ INT has a greater net EPA value because his opponent was near the endzone he was defending.