clock menu more-arrow no yes

Filed under:

QB Skill: A Narrative of Numbers (pt. 4)

New Orleans Saints v Philadelphia Eagles Photo by Mitchell Leff/Getty Images

Thanks to the nflFastR project and NFL NextGen Stats for the timely sources of data.


Welcome to the 4th and final installment of this numerical trilogy (RIP Douglas Adams).

Previously, I explained how EPA per Dropback (epa/d) is one of the best, most comprehensive measures of QB performance there is. I also showed that since 2017, Carson Wentz hasn’t done well in that stat. Of the 32 QBs with the most attempts in the last 3 seasons, Wentz’s epa/d of 0.01 ranks 28th.

The reason that is so important is that epa efficiency has about a 0.91 correlation to scoring. Wentz’s 28th-ranked epa efficiency predicts 1.79 points per drive(1) and whaddayouknow, he managed to just barely beat that with a 26th best 1.85 ppd.

So, Colts fans should care very much what his efficiency is in 2021 and should be. . . shall we say. . . concerned. But there is hype hope!

In part 3, I showed different ways in which passing stats can predict how a QB will play in the future and I believe there is good reason to think that Wentz will improve on his 3-year averages. So, I will address 5 reasons why Wentz will be better in 2021, attempting to quantify the efficiency impacts as I go.

Warning! Scientific wild-ass guesses ahead.


1) REGRESSION TOWARDS THE MEAN

I’m only half-joking when I say there is almost nowhere to go but up. QBs have up and down years for a myriad of reasons, and 2020 definitely looks like an outlier down year for Wentz.

Last year, he had arguably the worst performance of any QB. He had the league’s highest sack rate, 2nd worst accuracy, 3rd highest turnover rate, worst yardage efficiency, worst first down conversion rate, and 2nd worst epa efficiency, all of which caused him to be benched in week 13. All of those numbers are significantly lower than what he looked like in prior years.

As such, that performance may be biasing what I am estimating as his baseline average, but since I have limited data, this doesn’t give me many options. I could add in 2017, which is as much an outlier the other way and is really farther back than I want to go, or I could drop 2020. I chose the latter.

Therefore, “baseline” numbers from here on out are Wentz’s 2018-2019 averages.

I’ll talk about some of these in detail in the following sections, but right off the bat, you can see from the very last stat on the right, that I am raising his epa efficiency from 0.01 to his 2018-19 average of 0.08 (which still isn’t that good).


2) OFFENSIVE LINE

For these next few sections, I’m going to shift from epa to yards because that is how the items I will discuss are measured, but I’ll convert them back to epa.

Wentz’s yardage efficiency was 6.2 net yards per drop-back which ranked 24th. Since that number includes sacks and scrambles, O-Line protection is certainly a factor in the result and I think Wentz will see better numbers with the Colts in 2021. Here’s why.

Time to Decide (ttd) is the average time per drop-back that a QB takes to throw, scramble, or get sacked and Wentz had a slightly lower than average ttd to go along with his lower than average depth of target (adot). However, he faced slightly higher than average pressure (13th pr%) which suggests he wasn’t getting great protection. As a comparison, ESPN’s Pass Block Win Rate ranked the Eagles O-Line pass blocking 19th in both 2018 and 2019 so that lines up nicely with my numbers.

Time to Decide is strongly correlated to a QB’s combined sack + scramble rate so using that, I quantified the O-line impact by determining an expected sack + scramble rate(2). For Wentz, 2.67 seconds ttd equates to an expected sack + scramble rate of 9.9%, while his actual combined rate was 10.1% (scramble 4.0%, sack 6.1%). So, he bailed on 0.2% of passes more than expected.

Conversely, the Colts numbers point to far better protection. In the last 2 years, the average ttd for Colts QBs was 2.66 seconds, which translates to an expected 9.7% sck + scrm rate. The actual rate was 7.8%, beating expectations by 1.9%. The numbers were similar for both Brissett and Rivers across years and Pass Block Win Rate gives a weighted rank of 7th for those 2 Colts seasons, which again all agree nicely with each other.

Since Eric Fisher is a new piece and a bit of an unknown both in play quality and playtime, it is reasonable to temper those expectations. So, I am going to adjust the 1.9% advantage down to a flat 1.0% . . . because I can.

So, the 1.0% better than average numbers the Colts line gives and the 0.2% below average numbers the Eagles line gave, should result in a 1.2% reduction in Wentz’s sack/scramble rate. Assuming he maintains his 2.67 ttd as well as his historical ratio of sacks to scrambles, the revised data looks like this:

.

 dropbacks sck + scrm rate sck% scrm% sck scrm yds/sck yds/scrm yards
 dropbacks sck + scrm rate sck% scrm% sck scrm yds/sck yds/scrm yards
2018-19 Actual 1113 10.10% 6.10% 4.00% 68 44 -6.4 6.1 -163
Change -1.20% -0.70% -0.40% -8 -5 22
Revised 1113 8.90% 5.40% 3.50% 60 39 -6.4 6.1 -141

This results in a minor net yardage change of only 22 yards, but it shifts 8 sacks and 5 scrambles to 13 additional attempts, which will be captured in the next section.


3) RECEIVING YARDS

Pass length, in general, has an inverse relationship with yards after the catch (yac): the shorter the pass, the longer the yac and vice versa. However, while Wentz’s completions were on the shorter side (22nd ay/c), instead of producing a high-ranked yac, the Eagles receivers had the 6th shortest yac (27th).

This is very unusual and can have many causes like poor receivers (RBs and TEs included), good defenses, predictable play calling, poor blocking on screens, etc. For this analysis, the specific cause isn’t as important as the result, which is that there was poor yardage gain after the catch, which was an anchor on Wentz’s yardage efficiency (25th ypa).

The impact of this can be measured with YAC over expected (yacoe), which adjusts yac for pass length, yards to go, field position, and other variables. That measure shows the Eagles provided -0.5 fewer yac than expected. Conversely, the Colts receivers, over the last 2 seasons, have provided +0.3 yac more than expected, a swing of 0.8 additional yards per completion.

So, assuming Wentz maintains the same passing depth and completion rate and the Colts receivers can maintain their higher than expected yac, then his revised passing numbers would be:

.

 att cmp% cmp ay/c yac yds/c yards
 att cmp% cmp ay/c yac yds/c yards
2018-19 Actual 1001 66.23% 663 5.9 4.8 10.7 7061
Change 13 8 0.8 0.8 621
Revised 1014 66.23% 671 5.9 5.6 11.4 7682

Combining the 621 yards from receiver improvements and 22 yards from O-Line improvement, Wentz’s net yards per drop-back (ny/d) increases from 6.2 to 6.8, which is officially in the “not terrible” category.

I can use ny/d to estimate epa/d and Wentz’s 6.8 yardage efficiency equates to 0.104 epa efficiency(3), which is an improvement over his 0.08 baseline. To give some scope to this, 0.104 would have ranked 19th in epa/d last year.


4) OPPONENT DEFENSE

This is the part where the already guessy guesses get guessier.

In part 3, I showed that over the last 3 years, the Eagles had faced harder than average defenses. However, that included 2020. When looking only at his 2018-19 opponents, Wentz had it a bit easier than the average QB.

Opponent pass defense (opd) is the average amount of epa/d a team’s opponents gave up in the games prior to their meeting (and excluding any previous meetings). For Wentz, his 22nd ranked opponents gave up 0.07 epa/d which is about 0.01 more than the average team’s opponents.

In other words, Wentz had it 0.01 epa easier than average. However, if I assume all defenses look exactly the same next year, then the 2021 Colts opponents will give up 0.035 more epa/d than average, so Wentz should have it even easier: specifically about 0.025 epa/d easier (.035 - .010).

However, some of the adjustments I have already made from O-Line and receiver improvements already came from an easier schedule and I don’t want to double count their impact. So I’m going to trim the 0.025 schedule advantage down to 0.01, which will increase his revised epa/d to 0.114.


5) RUN GAME

This is the guessiest of all.

Wentz was on a “pass-first” team that did not have a very successful run game, as opposed to the Colts which have been the opposite the past few years.

The Eagles had a 54% early down pass rate(4) making them the 10th most likely to throw on early downs. Their run game had a 22nd ranked weighted rushing success rate (wrsr) showing they were less than average at getting TDs, first downs, or chunk yardage out of their runs.

These inputs have been strongly correlated with passing efficiency and so moving to a run-first, successful run game team should improve Wentz’s numbers. Over 2019-20, the Colts had the 27th ranked early down pass rate and the 4th highest rushing success rate, so the improvement should be significant.

Using ed% and wrsr as inputs, I came up with a 0.026 improvement for Wentz in 2021(5). However, inherent in that number is an assumption that ed% and wrsr cause epa/d and not the other way around. In other words, does good passing cause good running or vice versa? The answer is probably both as they are intrinsically related. So, I’ll cut the 0.026 in half and bump up his revised epa/d by .013 to a final value of 0.127.


THE WHOLE PICTURE

The vast majority of the articles that I write are about quantifying the past. I try to talk about how well a QB or team actually performed especially in the cases where the scoreboard doesn’t agree and the conventional wisdom is just plain wrong. Those numbers and conclusions come with a high degree of confidence.

However, when the analytical lens is flipped to look at the future, then the answers become unfocused . . . okay that metaphor was clunky, but you get the idea. The point is that predicting the past is easy and predicting the future is hard.

I do believe the Colts have a better O-line, better run game, better receiving options/scheme, and easier opponents than the Eagles did in 2018-19, so I absolutely believe Wentz’s efficiency will be much better in 2021. However, there is about a 0% chance that it will manifest itself in precisely the way I have laid out.

I predicted a 0.127 epa/d in total and I think that is certainly achievable, regardless of how likely I think it is. In 2020, that would have ranked 14th and equated to an expected ppd of 2.21. Therefore, even with drastic improvement from Wentz, that is a 0.26 ppd lower than the Colts scored in 2020 (2.47 ppd, 11th).

The Colts defense held opponents to 1.91 ppd last year, which was the 6th lowest. If they have to make up that missing 0.26 ppd, then they will have to improve to a top 3 defense. Only then can we expect a similar win total.

So, in summation:

  • I think Wentz will improve.
  • I think he can get to about 0.13 epa/d.
  • I think that is not enough, by itself, to get us to 11 wins.
  • I think all of these projections may be the delusional ravings of a lunatic mind.

FOOTNOTES

1) Based on linear regression of epa/d against points per drive since 2011 (0.91 correlation, 0.83 R-squared).

ppd =3.475 * epa/d +1.768

Points per drive is defined as all offensive points (including extra points) less any safeties and opponent turnover returns for touchdowns divided by the number of drives excluding drives that end due to time expiration.

2) Based on linear regression of ttd against sck = scr rate since 2018 (0.73 correlation, 053 R-squared)

sck + scr rate = 0.1475 * ttd - 0.2946

3) Based on linear regression of epa/d against net yards per dropback since 2011 (0.88 correlation, 0.78 R-squared).

epa/d = 0.1519 *ny/d - 0.925.

This conversion assumes that non-yardage events (TDs, INTs, first downs) are tied directly to ny/d. In other words, it assumes if a QB has the 5th best ny/d, then he also has the 5th best TD rate, First down rate, etc. This isn’t specifically true with Wentz, but his numbers were fairly aligned, so I’ll allow it.

4) 1st or 2nd downs in game neutral situations with > 2 yards to go. Game neutral is defined as between 20% and 80% win probability excluding the last 2 minutes of each half.

5) Based on linear regression of epa/d against ed% and wrsr since 2011 (0.39 R-squared).

epa/d = 1.27 * wrsrs + 0.72* ed% - 0.72