Following up on Seahawks run/pass season splits

In his Seahawks midseason awards, Danny Kelly made an encouraging observation:

Wilson's second half splits have by-and-large been better than
his first halves, with completion percentage, yards per attempt,
and rating improving every year after the halfway point

That's good as far as it goes, but those statistical numbers are not independent (more completions causes higher yards/attempt and a higher passer rating) and they don't account for rushing or sacks. So I put together some more aggregate performance measures. In the process, I made another surprising discovery about the Seahawks' season splits.

The Value of Aggregate Statistics

Last season, Wilson saw a big drop in his passing numbers. He also rushed for a remarkable 860 yards, more than the top running back on 18 other NFL teams.

A knee-jerk reaction might be that Wilson was struggling as a passer but had improved his running ability. And this is wrong. Most of Wilson's big runs were scrambles on pass plays, which means his passing numbers were down because he was good at recognizing opportunities where it was better to run. An opposition defense has to choose between a quarterback spy and an extra man covering wide receivers; they have to choose between a pass rush that contains the quarterback and one that applies maximum pressure. And you take what the defense gives you. It's no different than when a quarterback flips his #1 receiver in response to frequent double coverage; it would be silly to conclude that "He's gotten worse at throwing to Johnson but better at throwing to Smith."

Similarly, Wilson has been sacked more often than ever in 2015, but his yards/attempt is the highest for any season's first 8 games and his completion percentage is an absolute career high. These are not two separate, independent trends! By looking for better targets, Wilson takes more sacks and throws fewer incomplete passes.

An aggregate statistic like net yards per attempt will count sacks as passing plays (adding to the number of attempts and deducting the lost yards), so we can see whether or not the benefit of better passes is worth the sacks taken. Better is ANY/A, adjusted net yards per attempt, which includes a bonus for touchdowns (+20 yards) and a penalty for interceptions (-45 yards).

We can get rush yards into the same aggregate with TANY/A (total adjusted net yards per attempt), which simply counts quarterback runs (and rushing TD's) in the same pool as pass attempts.

all Seahawk statistics in this article do not include quarterback kneel-downs

Split	any/a	tany/a
2012 games 1-8	5.38	5.22
2012 games 9-16	8.86	8.78
2013 games 1-8	6.62	6.66
2013 games 9-16	7.59	7.59
2014 games 1-8	6.43	7.02
2014 games 9-16	7.03	7.52
2015	6.11	6.18

Danny's conclusion still holds, as I was sure it would.

Quarterback Burden and Play Frequency

But I haven't led you all this way for nothing. Another important factor is quarterback burden-- if Wilson drops back to pass more frequently, his total stats will improve, but his rate stats (like TANY/A) should be expected to drop as defenses call plays with less attention given to running backs.

WARNING! MATH!
My own clever invention to measure this is the Contribution Adjustment. This uses the adjusted net yards gained by the quarterback (passing or running) per team play, so that when he hands off the ball he earns zero yards and the play counts. We then calculate a geometric mean between the adjusted net yards per team play and TANY/A (total adjusted net yards per quarterback play). Finally, this number is normalized for the typical league average of 62% quarterback plays.

So, between two quarterbacks with identical rate stats, the QB with the larger burden will always have a higher Contribution-adjusted TANY/A; between two quarterbacks with identical total stats (per team play), the QB with higher rate stats will have better numbers; and a quarterback who runs/passes on exactly 62% of his team's offensive plays will have a contribution-adjusted TANY/A equal to his normal TANY/A for easy comparison.
/END HARD MATH

The numbers for Wilson with quarterback play percentage and adjustments:

Split	any/a	tany/a	Play %	Contribution- adjusted tany/a
'12 1-8	5.38	5.22	55%	4.92
'12 9-16	8.86	8.78	51%	8.00
'13 1-8	6.62	6.66	59%	6.49
'13 9-16	7.59	7.59	52%	6.98
'14 1-8	6.43	7.02	62%	7.04
'14 9-16	7.03	7.52	56%	7.16
2015	6.11	6.18	63%	6.21

The good news, as previously noted, is that Wilson's numbers have always improved in the second half of the season.

The bad news is that Wilson's any/a and tany/a are the lowest through eight games since his rookie year. The slate of opposing defenses (using DVOA) has been pretty average, but the Seahawks have played just 3 out of 8 at home, and just 5 1/2 games with Marshawn Lynch.

Also, the numbers are slightly better with the contribution adjustment. Seattle has handed the ball to a running back 187 times on 500 scrimmage plays, meaning Wilson is passing or running 62.6% of the time, a career high quarterback burden.

And that leads to our surprising discovery: Wilson's play percentage has dropped in the second half of the season every year. It's dropped a lot. The smallest drop was 4% during his rookie year, and the average is 5.3%. Is this a good thing? Let's take a closer look.

Season Rushing Splits

Every coach wants his team to run the ball effectively, none more so than Pete Carroll. If rush attempts (by running backs) increase, then, it could be an indicator that Seattle is unusually good at improving its running game throughout the course of the season.

Could be. Some other possible causes:

* The passing game gets worse and the Seahawks are forced to run

* All teams get better at running the football (or playing offense in general) throughout the course of the season, and this is not exceptional

* Seattle has more blowout wins and is running more in the fourth quarter to kill clock

* November & December weather forces the Seahawks to run more often

The depressing first option we can already rule out, because we've seen that Wilson's rate stats always improve in the second half of the season.

Now let's see what's "normal" for the league as a whole:

Split	Rush Att	Rush/game	Yards/run	TD%
2012 games 1-8	6688	26.1	4.37	2.9%
2012 games 9-16	6875	26.9	4.43	3.0%
2013 games 1-8	6623	25.9	4.25	3.0%
2013 games 9-16	6851	26.8	4.38	3.1%
2014 games 1-8	6698	26.2	4.28	2.9%
2014 games 9-16	6610	25.8	4.33	2.8%
Average games 1-8	6670	26.1	4.3	2.9%
Average games 9-16	6779	26.5	4.38	3.0%
Ave increase	109	0.4	0.08	0.03%

That's very raw data which counts quarterback runs, but it should still work for comparing the season split. Improvement in running looks consistent, but pretty small.

Now let's see how the Seahawks have performed, looking only at the running back carries (adjusted yards/attempt includes a 20-yard bonus for TD's):

Split	RB rush	RB yds	RB TD's	yards/att	TD %	adj y/a
2012 1-8	210	927	3	4.41	1.4%	4.70
2012 9-16	228	1168	9	5.12	3.9%	5.91
'12 playoffs	49	220	2	4.49	4.1%	5.31
2013 1-8	190	789	7	4.15	3.7%	4.89
2013 9-16	220	864	6	3.93	2.7%	4.47
'13 playoffs	79	385	4	4.87	5.1%	5.89
2014 1-8	182	795	6	4.37	3.3%	5.03
2014 9-16	225	1118	8	4.97	3.6%	5.68
'14 playoffs	71	374	2	5.27	2.8%	5.83
2015	187	810	3	4.33	1.6%	4.65

Seattle's tendency to improve the running game has been much greater than league average. Counting only the regular season splits, the Seahawks increased their yards/carry by +0.36 yards after the midpoint, more than four times the league average. Running in the playoffs tended to improve even further, and was always better than games 1-8, giving further evidence that this is a meaningful trend and not merely statistical noise.

Also, we can probably rule out weather and garbage time as factors in increasing rush attempts, because both have a strong tendency to reduce the overall yards/carry.

So, being afflicted with incurable Seahawk Optimism, I can only reach one conclusion: The combination of young offensive linemen and the difficulties of our zone blocking scheme consistently produces weaker performances in the early part of the year, with significant improvement to be expected as the season continues.