Defensive Statistics and Team Rankings

tims4wins · Nov 26, 2015

Aside from DVOA, which is more of a black box, I like to look at per drive statistics because they eliminate pace of play / # of possessions. So yards per drive and points per drive are pretty good stats.

The Pats are:
10th in yards per drive allowed (30.43); leader is Carolina at 24.98
5th in points per drive (1.60); leader is Carolina at 1.39
17th in turnovers per drive (.126); leader is NYG at 2.02
25th in plays per drive (6.03); leader is Houston at 5.31

What is interesting is that the Pats are 10th in yards per drive allowed despite being only 25th in plays per drive, which means they are playing pretty well on a per-play basis.

singaporesoxfan · Nov 26, 2015

At a basic level you have to first take out pick-sixes and safeties from points allowed in evaluating a defense. I'm guessing Denver slightly edges out the Patriots in points allowed once those are factored in.

amarshal2 · Nov 26, 2015

tims4wins said:
What is interesting is that the Pats are 10th in yards per drive allowed despite being only 25th in plays per drive, which means they are playing pretty well on a per-play basis.

I'm not sure I follow the "playing well on a per play basis" point you are making.

What I think it means is that the Pats give up first downs but not chunk plays. They'll let the other team run a bunch of plays and get first downs but they won't let them get a bunch of yards and they really won't let them get a bunch of points.

Bend but don't break.

singaporesoxfan said:
At a basic level you have to first take out pick-sixes and safeties from points allowed in evaluating a defense. I'm guessing Denver slightly edges out the Patriots in points allowed once those are factored in.

This is true. Of course you should also adjust for drive starting position but now we're getting complicated.

I came away from those drive stats impressed with Carolina. I didn't realize they had close to the best defense in the NFL.

tims4wins · Nov 26, 2015

amarshal2 said:
I'm not sure I follow the "playing well on a per play basis" point you are making.

What I think it means is that the Pats give up first downs but not chunk plays. They'll let the other team run a bunch of plays and get first downs but they won't let them get a bunch of yards and they really won't let them get a bunch of points.

Bend but don't break.

Right, pretty much this. They may give up some first downs but like you said generally force teams to use all three downs and don't give up huge chunks. Other teams run a good amount of plays but don't gain a ton of yards. Less three and outs but also less scoring.

Oppo · Nov 26, 2015

I hate the red zone scoring (td) %. It should be what % of the total red zone points possible do you score. A team could have a higher td % with turnovers and missed FGs and be considered a better red zone team than someone who scores slightly less td but has fewer turnovers and more FG.
(And the opposite to fit the thread title)

tims4wins · Nov 26, 2015

Oppo said:
I hate the red zone scoring (td) %. It should be what % of the total red zone points possible do you score. A team could have a higher td % with turnovers and missed FGs and be considered a better red zone team than someone who scores slightly less td but has fewer turnovers and more FG.
(And the opposite to fit the thread title)

Sometimes the telecasts will reference points per red zone possession. Far better stat.

jmcc5400 · Nov 26, 2015

For what it's worth, I quickly checked yesterday and roughly a third of the points given up by the Pats have come after they were already at a 99 % win expectancy. So there are some "soft" yards they have more or less willingly given in exchange for clock. Healthy, I think they clearly are a top 5 defense. (And I have a feeling they may be tired of hearing how good Denver's defense is)

mostman · Nov 26, 2015

jmcc5400 said:
For what it's worth, I quickly checked yesterday and roughly a third of the points given up by the Pats have come after they were already at a 99 % win expectancy. So there are some "soft" yards they have more or less willingly given in exchange for clock. Healthy, I think they clearly are a top 5 defense. (And I have a feeling they may be tired of hearing how good Denver's defense is)

That's a pretty interesting stat. Mind sharing where you pulled the data from?

jmcc5400 · Nov 26, 2015

Pro football reference boxscores

DJnVa · Nov 26, 2015

mostman said:
That's a pretty interesting stat. Mind sharing where you pulled the data from?

Yeah...all you have to do is look at the boxscore, and it gives a per minute win percentage. Find when it hits 99% or whatever threshold you want, and then check the PBP after that point.

lexrageorge · Nov 28, 2015

Some specifics on late game scores against the Patriots:

- Steelers scored a TD with 2 seconds left that made it 28-21 Patriots.
- The Bills scored 3 straight TD's after entering the 4th quarter down 37-13. Two of those came on relatively short fields set up by Pats turnovers (1 by downs, 1 via strip sack of Brady).
- Jaguars scored a late TD with 1:22 left to close the gap to 51-17.
- Redskins scored a TD with 25 seconds left to make it 27-10.

There are 2 late scores that made it one score games, with enough meaningful time for a potential tying score late:

- The Participation Trophy winners scored a TD with 1:19 left to make it 34-27.
- The Yets kicked a long FG with 18 seconds left to make it a 1 TD game.

That's 52 points scored when the game was essentially over. I wouldn't give the defense a pass on all of them, but clearly those points do not mean as much as those scored when the game is on the line early.

bankshot1 · Nov 28, 2015

jmcc5400 said:
For what it's worth, I quickly checked yesterday and roughly a third of the points given up by the Pats have come after they were already at a 99 % win expectancy. So there are some "soft" yards they have more or less willingly given in exchange for clock. Healthy, I think they clearly are a top 5 defense. (And I have a feeling they may be tired of hearing how good Denver's defense is)

Bingo.

you have to view defense and the stats in context.

If you think about the Pats early blow-out wins, they often had big leads early, which means the opponent almost has to give up the ground game and go to the air far earlier than they might like. So the Pats run defense might look better than it is, and the pass defense worse on a yards allowed basis. Also we've seen several TD's scored in garbage time. We as fans don't like to see the Pats scored on, but a last second TD to the Steelers in a 14-point game is pretty meaningless.

IMO a top 5 D is a reasonable conclusion.

bankshot1 · Nov 28, 2015

dcmissle · Nov 28, 2015

Well this week provides an excellent, fair test. 19 times out of 20, the Pats win this game, if they win it, via defense and special teams. I don't see the Pats completing extended drives with TDs, and I don't see Denver handing us the game. At the same time, Pats can't complain about bringing a knife to a gunfight considering who is playing QB for them.

GeorgeCostanza · Nov 28, 2015

bankshot1 said:
Bingo.

you have to view defense and the stats in context.

If you think about the Pats early blow-out wins, they often had big leads early, which means the opponent almost has to give up the ground game and go to the air far earlier than they might like. So the Pats run defense might look better than it is, and the pass defense worse on a yards allowed basis. Also we've seen several TD's scored in garbage time. We as fans don't like to see the Pats scored on, but a last second TD to the Steelers in a 14-point game is pretty meaningless.

IMO a top 5 D is a reasonable conclusion.

The post above yours nails what you're saying in detail, did you miss it?

Nice work Lex. I'm pleased to be surprised that the defense has gone above and beyond my expectations, starting with Butler. Look forward to seeing him battle either DT or manny. Hopefully Ryan can pull out an INT, I have a feeling they will need some short fields for the offense and maybe even a defensive score.

bankshot1 · Nov 28, 2015

GeorgeCostanza said:
The post above yours nails what you're saying in detail, did you miss it?

I hadn't read lex's post before I responded to jmcc's post.
And while my post does repeat lex's point about irrelevant late scores, my larger point was the "early lead" factor has led to deflated running yards and inflated passing yards yielded by the Pats defense reflecting a change in opponent game strategy when they're down 2-3 scores in the 2nd quarter.

Perhaps you missed it.

dbn · Nov 28, 2015

There seems to me an obvious, straightforward, easy-to-understand method to compare defenses. In fact, I'd be surprised if it isn't already being done and I'm just not aware of it. The only large problem (that I can think of) is that it doesn't naturally adjust for garbage time, though one could do something like add an inverse weight based on win expectancy.

I once asked a question in this forum that was answered (very well, thanks for whoever answered it, I forget) using point-scoring expectancy as a function of field position. So someone must have calculated that. If not, it shouldn't be difficult to calculate if one has the data.

So one can simply calculate, for each defense, the number of points per possession they were expected to give up based on drive starting positions, and compare that to the actual number. This leaves out something that I think is important, namely yards given up on drives that don't result in a score. As mentioned up-thread, it's the points that matter, but (in a non-linear way) the yards matter, too, because they affect the ability of your offense to score by the resulting field position. There is a way around this which I'll describe but spoiler, as this post will otherwise become quite unwieldily.

As simple example of how yards allowed matter, consider two defensive drives starting with 95 yards to give for a TD. Both have the same point-allowed expectation value. Say the first drive allows 40 yards and forces a punt that is downed at the opposite 5 yard line. Say the second drive is a 3-and-out that forces a punt that is downed close to mid-field. The offense of the team who's defense we are considering has a much higher point-scored expectation value in the latter situation compared to the former. In essence, the defense has spotted their offense a certain number of (prospective) points more in the latter compared to the former, because they prevented more yards, for which they should be credited.

Here is how you do it. In the same manner that "points-scored expectation value as a function of starting yard line" (let's call that E(X), where X = yard line at which drive starts), you calculate the following functions. Fraction of drives that end in a field goal based on starting position, f(X), the same for touchdowns, t(X), the same for safeties, s(X), the same for pick-6s, p(X), and finally the fraction that end in non-scores, n(X). By construction:
f(X) + t(X) + s(X) + p(X) + n(X) = 1.

One more thing we'll need is the yard line at which a drive that starts at X, and does not lead to a score, resultantly starts for the opposing team: Y(X); i.e., after a punt or unsuccessful 4th-down attempt.

Now, clearly,
E(X) = 7 x t(X) + 3 x f(x) - 2 x s(X) - 7 x p(X).

But more interesting is the difference between points expected to be given up based on X, and the points expected to be gained by the offense on the subsequent drive based on X. We can define that as

E*(X) = E(X) - n(X) x E(Y(X)).

You don't even need to divide by number of drives, as it already considers drives in pairs of O/D. An obvious flaw is that, when comparing expected to actual, one needs to include the result of the punt return in the actual, but if one considers the punt return as part of the D, it isn't a problem. Alternately, one could use a "expected starting position based on a punt being forced at yard-line Z" in place.

One final thought is about how those functions are calculated. If you just use the data, there will be noise enough that, e.g., the computed touchdown % might be higher starting at the 24 compared to starting at the 25. We don't want that as it isn't sensible. So you should either do a smoothing of the data or, better yet, find a functional form such as a line, power-law, exponential, etc., and do, e.g., a maximum-likelihood fit to the raw data, and then used that parameterized description of the data for the functional forms.

Unless I'm either missing something stupid, or this has already been done, someone should do this if they have the data and the time. Or, if you have the data and would like to see it done, PM me; maybe I'd find time to do it.

Anyhow, thoughts?

williams_482 · Nov 28, 2015

dbn said:
There seems to me an obvious, straightforward, easy-to-understand method to compare defenses. In fact, I'd be surprised if it isn't already being done and I'm just not aware of it. The only large problem (that I can think of) is that it doesn't naturally adjust for garbage time, though one could do something like add an inverse weight based on win expectancy.

I once asked a question in this forum that was answered (very well, thanks for whoever answered it, I forget) using point-scoring expectancy as a function of field position. So someone must have calculated that. If not, it shouldn't be difficult to calculate if one has the data.

So one can simply calculate, for each defense, the number of points per possession they were expected to give up based on drive starting positions, and compare that to the actual number. This leaves out something that I think is important, namely yards given up on drives that don't result in a score. As mentioned up-thread, it's the points that matter, but (in a non-linear way) the yards matter, too, because they affect the ability of your offense to score by the resulting field position. There is a way around this which I'll describe but spoiler, as this post will otherwise become quite unwieldily.

As simple example of how yards allowed matter, consider two defensive drives starting with 95 yards to give for a TD. Both have the same point-allowed expectation value. Say the first drive allows 40 yards and forces a punt that is downed at the opposite 5 yard line. Say the second drive is a 3-and-out that forces a punt that is downed close to mid-field. The offense of the team who's defense we are considering has a much higher point-scored expectation value in the latter situation compared to the former. In essence, the defense has spotted their offense a certain number of (prospective) points more in the latter compared to the former, because they prevented more yards, for which they should be credited.

Here is how you do it. In the same manner that "points-scored expectation value as a function of starting yard line" (let's call that E(X), where X = yard line at which drive starts), you calculate the following functions. Fraction of drives that end in a field goal based on starting position, f(X), the same for touchdowns, t(X), the same for safeties, s(X), the same for pick-6s, p(X), and finally the fraction that end in non-scores, n(X). By construction:
f(X) + t(X) + s(X) + p(X) + n(X) = 1.

One more thing we'll need is the yard line at which a drive that starts at X, and does not lead to a score, resultantly starts for the opposing team: Y(X); i.e., after a punt or unsuccessful 4th-down attempt.

Now, clearly,
E(X) = 7 x t(X) + 3 x f(x) - 2 x s(X) - 7 x p(X).

But more interesting is the difference between points expected to be given up based on X, and the points expected to be gained by the offense on the subsequent drive based on X. We can define that as

E*(X) = E(X) - n(X) x E(Y(X)).

You don't even need to divide by number of drives, as it already considers drives in pairs of O/D. An obvious flaw is that, when comparing expected to actual, one needs to include the result of the punt return in the actual, but if one considers the punt return as part of the D, it isn't a problem. Alternately, one could use a "expected starting position based on a punt being forced at yard-line Z" in place.

One final thought is about how those functions are calculated. If you just use the data, there will be noise enough that, e.g., the computed touchdown % might be higher starting at the 24 compared to starting at the 25. We don't want that as it isn't sensible. So you should either do a smoothing of the data or, better yet, find a functional form such as a line, power-law, exponential, etc., and do, e.g., a maximum-likelihood fit to the raw data, and then used that parameterized description of the data for the functional forms.

Unless I'm either missing something stupid, or this has already been done, someone should do this if they have the data and the time. Or, if you have the data and would like to see it done, PM me; maybe I'd find time to do it.

Anyhow, thoughts?

Advanced Football Analytics had an expected points model which was pretty much what you describe. Unfortunately the site is now defunct, and although the stats may have been moved to ESPN I can't seem to find them.

Devizier · Nov 29, 2015

I don't know how predictive it would be, but a WPA statistic might be informative. Since there's already PBP win probability numbers available, it would be pretty easy to generate. You could then make opponent adjustments post-hoc, if necessary.

singaporesoxfan · Nov 29, 2015

williams_482 said:
Advanced Football Analytics had an expected points model which was pretty much what you describe. Unfortunately the site is now defunct, and although the stats may have been moved to ESPN I can't seem to find them.

I met Brian Burke formerly of AFA at a talk recently. Unfortunately as part of his move to ESPN all his stats have be one proprietary. Part of a general trend that makes football statistical analysis harder than baseball (even outside of the inherent differences in the sports).

But yes, expected points is what you need to look for. Then you might want to adjust for strength of schedule - for example, I think the Pats have a good defense this year but it's also been made to look better by playing the AFC South and NFC East. On top of that since weather affects play so much I'd love to see something equivalent to park effects in baseball.

Super Nomario · Nov 29, 2015

Expected points is reported by Pro-Football-Reference:

http://www.pro-football-reference.com/years/2015/opp.htm

Pats are only 15th by this metric, interestingly. Some of that is they haven't gotten a ton of value out of turnovers - they're 23rd in EPA from turnovers. They're one of seven teams that doesn't have a turnover returned for a TD this year.

BaseballJones · Dec 14, 2015

Currently, the Pats rank 8th in the NFL in points allowed (which isn't totally fair, given the 3 TDs they gave up last week on special teams and the pick-six, but I guess other teams have those happen throughout the season). Other per-game rankings at the moment:

#6 in yards allowed (329.5)
#7 in pass yards allowed (230.8)
#10 in rush yards allowed (98.6)
#8 in pass completion allowed (59.4%)
#7 in yards per pass attempt allowed (6.8)
#2 in sacks (42)
#9 in opponent passer rating (83.1)
#11 in third down conversions (37.2%)
#10 in DSRS (pro-football-reference.com) at 2.6

A completely legitimate top 10 defense, with no glaring weakness. Solid against the run. Solid against the pass. They're not doing it with tons of turnovers either (just 18th in the NFL with 17 takeaways). So they're just playing really good overall defense.

I wish they were generating more turnovers, but I love that they're not relying on turnovers (which can be a bit of a random thing at times) to stop opposing offenses, but rather that they can get the job done without them.

bakahump · Dec 15, 2015

(Almost) Time to give BB and MPat their due.

They built the defense (or at least appeared to) based on a better Pass Rush countering weaker secondary personnel.

They have (again thus far..) been right.

Jungleland · Dec 15, 2015

Impressive stuff, and backed by the eye test. Though they have faced what seems an easier than usual slate of quarterbacks this year, they have done exceptionally well containing some of the league's better receivers in Beckham, Hopkins, and Thomas in recent games.

Where I get excited is that several guys have taken big leaps of late - Logan Ryan, the defensive line, Patrick Chung in particular - and almost entirely in the games since Collins first went out. If those improvements are real one has to think the ceiling for this defense is (significantly?) higher when the starting linebacker tandem is finally back on the field together. Shame that McCourty's injury means we likely won't see the full strength unit until January, but there seems to be a lot of room for improvement over these already good numbers if/when everyone's healthy.

Super Nomario · Dec 17, 2015

For years we had the bend-but-don't break D, that would allow long drives but then get a big step or hold tough in the red zone.

This year? The Pats are #1 in forcing three-and-outs, with 51. The Jets are #2 with 47, Denver #3 with 42.

McBride11 · Dec 17, 2015

Super Nomario said:
For years we had the bend-but-don't break D, that would allow long drives but then get a big step or hold tough in the red zone.

This year? The Pats are #1 in forcing three-and-outs, with 51. The Jets are #2 with 47, Denver #3 with 42.

The game threads disagree with this fact.

It is pretty interesting how far this D has come. The D line, with the number of high draft picks l, makes sense. Plus Sheard being higher profile pickup. Ryan and Butler have been huge surprises and really make the D imo. Imagine having to spend on Revis or whatever 4th CB we would be playing now.

Defensive Statistics and Team Rankings

Member

PN23's replacement

Well-Known Member

Member

PN23's replacement

Member

PN23's replacement

Member

Member

Member

Dorito Dawg

Member

Member

Member

Deflatigator

tiger king

Member

Member

Member

Member

Well-Known Member

Member

ivanvamp

Well-Known Member

Member

Member

Member