Creating better football stats

[sabotai]

One of the things with football I've been thinking about is that the stats are fairly simplistic. Completion percentage for QBs treat screen and shuffle passes the same as passes that go 15-20 yards down field and YPA give the QB credit for the receiver's YAC. Rushing stats treat a 15 yard draw play on 3rd and 20 the same as a 2 yard run on 3rd and 1. A defense gets the blame (points against) for a FG following the offense coughing up a fumble on their own 20 yard line.

So I've been brainstorming lately on some ideas for better statistics. Here's a few of them that I've looked at and would like some help from you guys picking apart my ideas, showing me how awful they are and shaming me back into retirement.

The first two I've only used on 9 QBs from 2012 so far.

Expected Completion Percentage (ECP)

This is based on the idea that not all passes are created equal. A QB that throws 100 screen passes should complete X amount of them, while a QB that throws 100 passes 20 yards down field is obviously going to complete less than X of them.

The idea is that the field is broken into 4 zones. At or behind the line of scrimmage (Z1), 1-7 yards down field (Z2), 8-14 yards down field (Z3) and 15+ yards down field (Z4). In 2012, 16.47% of all passes were in Z1, 40.99% in Z2, 21.15% in Z3 and 21.33% in Z4. My initial thought was to have a 25/25/25/25 split, but that would have made one of the zones extremely small. As you can see, quite a lot of passes fall in the 1-7 zone. If I did the 25s split, one of the zone would have been 3-6. That's a bit too tight.

League average completion percentage per zone
Z1: 74.36%
Z2: 68.35%
Z3: 58.00%
Z4: 41.20%
All passes: 61.33%

So what I do is take the individual QBs distribution of passes and, based on the league average completion for each zone, figure out what the QB's completion percentage would be if he was completely average.

Let's take Andrew Luck in 2012. He had a completion percentage of 54.07% (official number). However, his pass distribution was 12.94% in Z1, 35.28% in Z2, 23.62% in Z3 and 27.99% in Z4. He was throwing to targets in Z1 and Z2 below league average, and to targets in Z3 and Z4 above league average, so it's expected that his completion percentage would suffer. But by how much?

If Andrew Luck was completely average, that is, if he hit his targets along the league average, what would his completion percentage look like?

Total Attempts: 618*
Z1: 80 attempts - at 74.36% - 59.488 completions
Z2: 218 attempts - 68.35% - 148.24 completions
Z3: 146 attempts - 58.00% - 84.68 completions
Z4: 173 attempts - 41.20% - 71.276 completions
Total Expected Completions: 363.684 (Actual: 340**)
Expected Completion Percentage: 58.85% (Actual: 54.07%***)

Still well below average, but not as bad as 54.07% vs. 61.33% (league average).

* - Pro-Football-Reference has his total attempts at 627. I'm using the football-outsider's spreadsheet and I'm not sure where those 9 passes went. For all of the QBs, my counts are usually ~1-3 off with Luck's number being more to the extreme. When I do the count, I'm counting passes that are marked as "passes" in the playtype field in the spreadsheet, and have the result of either "complete", "incomplete" or "interception". I'm guessing those 9 passes were some type of broken play or trickeration that got marked something different on the spreadsheet. I'd have to track them down to find out what happened.

** 339 officially, odd because I'm simply counting plays marked "passes" and results marked "complete". But I'm guessing slight differences in numbers are just the reality of using a spreadsheet like this without investigating each and every discrepancy.

*** 340/618 = 55.01%. The spreadsheet discrepancies add up. Something I'm really noticing as I write this up.

The idea for this stat was to find a better benchmark to compare the QB's completion percentage to than the league's overall average. IOW, two QBs who complete exactly 61% of their passes are not necessarily equal. One QB's 61% might actually be better than average while another's could actually be below average

The 9 QBs list of expected completion percentage vs. their actual cmp% (according to my spreadsheet, not official numbers)
League Average CMP%: 61.33%
Tom Brady: 61.26% ECP - 63.55% (+2.29%)
Peyton Manning: 61.29% ECP - 69.02% (+7.73%)
Eli Manning: 60.31% ECP - 60.11% (-0.20%)
Robert Griffin III: 61.89% ECP - 66.58% (+4.69%)
Matthew Stafford: 60.98% ECP - 60.17% (-0.81%)
Drew Brees: 61.48% ECP - 63.65% (+2.17%)
Tony Romo: 61.28% ECP - 65.94% (+4.66%)
Andrew Luck: 58.85% ECP - 55.02% (-3.83%)
Matt Ryan: 61.97% ECP - 68.62% (+6.65%)

Distance Adjusted Passing Completion (DAPC)

A different way of measuring pretty much the same as above.

Taking the pass distributions the same as above, what I do here is take the QB's completion percentage in each zone, and apply to the league average for attempts in each zone.

This time I'll use Eli Manning as the example

Eli's pass distribution + completion percentage by zone
Total attempts: 534 (official numbers 536)
Z1: 56 attempts, 10.49% - 39 completes, 69.64%
Z2: 214 attempts, 40.07% - 142 completes, 66.36%
Z3: 151 attempts, 28.28% - 93 completes, 61.59%
Z4: 113 attempts, 21.33% - 47 completes, 41.49%
Total completions: 321 - 60.11%

(59.89% completion officially - 321/536 = 0.5989; 321/534 = 0.6011)

Eli completes passes in zones 3 and 4 at a higher rate than league average, but in zones 1 and 2 below league average. His distribution in zones 2 and 4 are along league averages, but he throws to zone 1 the least of any QB I've looked at so far, and the throws to zone 3 the highest.

Now, if we redistribute his attempts based on league averages, and then apply his completion percentage per zone...

Z1: 103.93 atts, 69.64% = 61.25 cmps
Z2: 218.89 atts, 66.36% = 145.24 cmps
Z3: 112.92 atts, 61.59% = 69.56 cmps
Z4: 113.90 atts, 41.59% = 47.38 cmps
Total completions: 323.43 / 534 = 60.57% DAPC

Here is the rundown of the 9 QBs I've done this for (using the completion pct. I get from the spreadsheet)

QB - Actual Cmp% - DAPC

Tom Brady: 63.55% - 63.52% DAPC
Peyton Manning: 69.02% - 68.50% DAPC
Eli Manning: 60.11% - 60.57% DAPC
Robert Griffin III: 66.58% - 65.39% DAPC
Matthew Stafford: 60.17% - 60.12% DAPC
Drew Brees: 63.65% - 64.39% DAPC
Tony Romo: 65.94% - 65.30% DAPC
Andrew Luck: 55.02% - 57.43% DAPC
Matt Ryan: 68.62% - 67.76% DAPC

Not really sure what information this number provides, though. QBs who throw deep more often generally benefit, QBs who throw short more than average get penalized. I was trying to find a way to get a better grasp of a QB's accuracy.

P.Manning's number takes a hit because it takes passes from his highest zone (Z2, 44.75% of his passes) and moves it to Zones 1, 3 and 4. He was actually below league average in passes to Zone 1. He completed 79.23% of his passes to Zone 2 (+10.88% to league average), and was 7-8% higher in zones 3 and 4. So taking passes out of Zone 2 and putting them in Zone 3 and 4 still hurt his number.

What I find most interesting after doing this is the differences in the QB's completion percentage per zone.

Eli completes passes at 3.59% above league average in the zone 3, but 1.99% below average in zone 2.
Peyton Manning demolishes league averages in zones 2-4, but is below league average in zone 1.
Matthew Stafford is within +/- 1.5% of league average in zones 1, 2 and 4 and is 6.39% below average in zone 3.
Drew Brees starts off ine Zone 1 at 7.69% below league average, is 1.68% below league average in zone 2, jumps up to 8.26% above league average in zone 3 and is 10.71% above average in zone 4.
Tony Romo is 1.15% below average in zone 1, 3-4% higher in zones 3 and 4, but is 6.65% higher in zone 2.
Matt Ryan, 1.76% above league average in zone 2, but is way above it in zones 1 (+14.43%), 3 (+8.96%) and zone 4 (+6.77%).

But then again, a lot of those percentages could swing widely because of sample sizes....

For the two above, a few ideas I had..

For one, I wanted to eliminate all screens and shuffle passes, but FO made a deal with ESPN that eliminated that information from the spreadsheets they provide (along with a few others I could have used)....so that's out.

I also thought about eliminating all passes that go either 5 or more yards behind the line of scrimmage (~2% of all passes - at that point, it's usually a broken play - not a test of a QB's accuracy) and eliminating all passes that go beyond 35-40 yards passed the LOS (~2-3% of passes). After a certain distance, a QB is just throwing deep. A hail mary isn't really a type of pass that gives an indication of a QB's accuracy.

I also thought about splitting them between red zone and non-red zone passes.

Giving QBs credit for a completion if the play was marked as dropped by the receiver.

I also thought of instead of zones, I'd give each pass distance a point value based on league completion percentage for that distance (smoothed out a bit due to sample size) to make it more of a curve. Passes 5 yards down field could be worth 0.94, 8 yards down field 1.00, 15 yards down field 1.195, etc. until 30 where it'd be 2 points, but then, like the idea above, not count passes over 40. (30-39 would all equal 2 points because a pass to the 30 yard line completes ~30% of the time, and a pass to the 39 yard line completes at ~30% as well - after 40 it starts dropping again until you get to the 50s and 60s where it becomes erratic due to sample size). And then just take that number and divide by attempts.

Moving on....

Points For and Against: Credit and Blame

I haven't test this one out much. I only looked at the Giants' first several games.

Here's the basic rules

If the offense or special teams gives the ball to the opponent's within their own 40 yard line, and they then score a FG, that FG gets attributed to the offense or ST. If they scored a TD, then 3 points against the offense or ST, and 4 against the defense (5 in the case of a convert 2pt). After all, the ultimate goal of the defense is to keep the ball out of the endzone. So, even if they get it at the 1, it's still up to the defense to keep them out.
Obviously pick 6's go against the offense, and returns for TDs go against ST.

If the ST or defense gives the ball to the offense within the opponents 40, the same as above. A FG is credited to the side that put the ball inside FG range, and if a TD is scored, the offense gets 4 points (5 in the case of a converted 2pt), the side that put the ball inside the 40 gets 3.

After several Giants' games, here is how the credit and blame was given. Although, I was using slightly different rules so I would have to redo them for the above (some rules gave 1.5 blame and credit for FGs, so that's why there are some .5s in the numbers below), but the general idea is this...

Points Against
Total: 209 points against
Offense Blame: 42.5 points (20.33% of points scored)
Defense Blame: 139 points (66.51% of points scored)
Special Teams Blame: 27.5 points (13.16% of points scored)

Points For
Total: 103 points for
Offense Credit: 95.5 points (92.72% of points scored)
Defense Credit: 7.5 points (7.28% of points scored)
Special Teams: 0 points (big fat ass zero % of points scored)

Not having done this for other teams, I have no idea how those numbers would stack up against league averages and whatnot.

Other ideas

Trying to go beyond the success/non-success for the running game. Maybe come up with a kind of "slugging percentage" for running and/or passing.

Anyway, I'd appreciate it if you guys picked these apart and showed me why these ideas are terrible and/or suggest ways to make them much better.

[Young Drachma]

Oohh...

[NobodyHere]

My head hurts

[Young Drachma]

So what's the % difference between the ECP and actual CP called? To me, that's the interesting part of that stat, is finding out someone who wildly outplays his expectation. Also, the stat is great for starters but for a system QB or backups, it might not be as useful a measure of anything other than anomalous performance. That's just blood spatter though, as I like where you're trying to go with these.

Will have to look at the others more closely.