Vela: A. Jones, CB Rankings and Ken Hamlin’s Best Position: K.C. Joyner Returns, Pt.1

[percyhoward]

Say Team A averages 20 offensive points on 12 possessions per game, while Team B averages 18 offensive points on nine possessions per game. Now put them in the same game, give them each the same number of possessions and have the offense perform at exactly their average level. Who wins? Team B, of course. If they each have nine possessions, Team B wins 18-15. If they each have 12, Team B wins 24-20.

This last part is what I'm talking about.

Are there really more points scored in games that have more possessions?

If an entire stat is going to be based on that assumption, first find out if the assumption is accurate.

 

[percyhoward]

Intuitively speaking, why shouldn't an increase in the number of attempts to score lead to more scoring?

As ridiculous as it may sound, an increase in your number of possessions doesn't mean an increase in your attempts to score, or your chances to score.

I never heard any coach say, "we've got to limit their number of possesions."

 

[Temo]

As ridiculous as it may sound, an increase in your number of possessions doesn't mean an increase in your attempts to score, or your chances to score.

I never heard any coach say, "we've got to limit their number of possesions."

Sure they do... that's what they mean by "keep the ball out of their hands" and "keep their offense (and our defense) off the field".

But besides that, the points per possession is meant more to compare defense vs. defense and offense vs. offense than anything. It's a way to measure a defense that gives up 14 points on 6 possessions vs. one that gives up 14 points on 10 possessions. I would contend that it's the same reason that yards per attempt is used as a stat... as well as ERA in baseball, etc.

 

[percyhoward]

Sure they do... that's what they mean by "keep the ball out of their hands" and "keep their offense (and our defense) off the field".

They're talking about TOP, or plays, maybe, but not possessions. The more three-and-outs a defense forces, the more possessions it creates for the opponent.

But besides that, the points per possession is meant more to compare defense vs. defense and offense vs. offense than anything. It's a way to measure a defense that gives up 14 points on 6 possessions vs. one that gives up 14 points on 10 possessions. I would contend that it's the same reason that yards per attempt is used as a stat... as well as ERA in baseball, etc.

But again, why distinguish? Why is a defense that gives up 14 on 6 possessions worse (on average) than a defense that gives up 14 on 10? If scoring doesn't go up as number of possessions increase (game-to-game), then what handicap did that 2nd defense have to overcome?

 

[Temo]

They're talking about TOP, or plays, maybe, but not possessions. The more three-and-outs a defense forces, the more possessions it creates for the opponent.


But again, why distinguish? Why is a defense that gives up 14 on 6 possessions worse (on average) than a defense that gives up 14 on 10? If scoring doesn't go up as number of possessions increase (game-to-game), then what handicap did that 2nd defense have to overcome?

I was just going to leave it at that, but I didn't want to make it seem like I was dodging.
I'm sure that scoring will increase with more possessions in games. However, at this point without bringing math into it, it's just talk. And seeing as how I don't have the inclination at this point to go into that sphere of analysis, I'll just respectfully agree to disagree for now.

 

[theogt]

so you propose a simple univariate comparison for what is admittedly a complex, multifaceted relationship?

You're unnecessarily adding complexity because you're looking for something else. You want a measure of the TEAM. We're looking for a measure of the DEFENSE.

 

[AdamJT13]

Not necessarily true? It also isn't necessarily false.

Like I said, the offense got the ball more often, and it was up to the offense what it did with it. If the offense ran fewer plays on 15 possessions than it did on eight possessions, that would be the offense's fault, not its own defense's fault.

Take your DVO stats from 2007 -- the correlation between offense and defensive numbers is -.26 (higher offense means better lower defense means better). Using that metric there is a relationship. Not that I am endorsing the metric

Is that a negative .26 correlation between a good offense and a good defense, or did you mean to say something else?

Using the DVOA rankings for 2007, there was a 0.30 correlation between offensive and defensive rankings. For 2006, it was a negative 0.08 correlation. For 2005, it was 0.32. For 2004, it was negative 0.11.

I can keep going, if you like, but I see very little overall correlation between having a good offense and having a good defense (measured by DVOA). What little correlation there is (which obviously doesn't equate to causation) probably can be explained by the fact that good teams try to load up for runs at the Super Bowl, while bad teams try to rebuild. So on average, teams tend to gravitate slightly more toward being good or bad on both sides of the ball, rather than good at one and bad at the other. But it's only slightly (as those correlations above seem to show).

I just picked out the handful of games where i remembered the worst performances. Just an example.

Worst defense or worst offense? Because like I said, the offense DID produce in several of those games.

If sacks go up it stands to reason that general pressure on the QB goes up -- that can lead to more throw aways, rushed throws that drop incomplete, and even a reliance on short quick passing. There's one explanation for you

So why did our opponents throw deeper more often last season?

 

[AdamJT13]

On average, that's exactly what happens, though.

Obviously, there's no guarantee that it will happen, but we're not talking about certainties, anyway. It's about the assumption that, in games with more possessions, more points will be scored--ON AVERAGE.

If there's no evidence that scoring goes up in proportion to possessions, that it doesn't make a whole lot of sense to factor number of possessions into a scoring stat.

If the offense's performance level is the same, scoring will go up as the number of possessions goes up. That's an inherent fact.

The problem with what you're asking is that the number of possessions within a game is often inversely related to performance -- the better you do on offense, the fewer possessions you'll usually have, while the better you do on defense, the more possessions your defense will face. The opposite also can be true, if there are quick scores (say, three plays or less), but those are more rare than the combined three-and-outs and quick turnovers.

So, generally, if there are a very small number of possessions in a game for each team, it's likely that both offenses were quite successful at sustaining drives (or that one was EXTREMELY successful). If there are a very high number of possessions, it's likely that both teams were fairly unsuccessful at sustaining drives (or that one was EXTREMELY unsuccessful at it). For anything in between, it could be any combination -- one team could be very successful and the other very unsuccessful, or both could be completely average. Regardless, when a team gets the ball more times, it SHOULD compile higher stat totals (total points, total yards and total time of possession). If it doesn't, then it's not performing like it should.

I looked at the defensive points allowed in each game for 10 different teams last season -- all four NFC East teams, the top two teams in points allowed, the middle two teams in points allowed and the bottom two teams in points allowed -- and divided it into how many legitimate possessions each defense faced in each game (eliminating all drives when the opponent either didn't have time or didn't intend to score). When the opponent had eight legitimate possessions, those defenses allowed an average of 18.08 points per game. With nine possessions, it was 19.07. With 10, it was 19.19. And with 11, it was 19.93. So yes, it appears that within a typical range of possessions, more possessions do mean more points, on average.

 

[percyhoward]

With nine possessions, it was 19.07. With 10, it was 19.19. And with 11, it was 19.93.

This is what I mean. According to per-possession stats, the average score for 11 possessions should be up over 23 points, not 19.93.

And your sample was of well over 100 games.

Your whole argument is based on the assumption that a team's points allowed per possession stays about the same as possessions are either added, or taken away. Remember?

If they each have nine possessions, Team B wins 18-15. If they each have 12, Team B wins 24-20.

Your recent research does nothing to support this...

9 possessions
points allowed per possession: 2.11 (19.07 total)

10 possessions
your assumption: 2.11 (21.1 total)
actual number: 1.92 (19.2 total)
The jump up from 19.07 points to 19.2 is minoscule. That extra possession, based on your assumption, should have put the opposing offense up over 21 points. They ended up not even cracking 20!

11 possessions
your assumption: 2.11 (23.21 total)
actual number: 1.81 (19.91 total)
This isn't going where you thought it would, is it? Missed by more than a FG here.

12 possessions
your assumption: 2.11
You didn't provide the actual number, which (correct me if need be) means it isn't anywhere near what you assumed it would be. Did the downward trend continue, or did the point totals suddenly skyrocket? Either way, you haven't shown that the amount of points a team allows per possession remains anything near constant when more possessions are added.

 

[AbeBeta]

Is that a negative .26 correlation between a good offense and a good defense, or did you mean to say something else?

It is negative -- which is why I wrote -.26 (note this is for DVOA not weighted DVOA which gives more weight to games later in the season, an adjustment that is irrelevant to the present discussion).

Using the DVOA rankings for 2007, there was a 0.30 correlation between offensive and defensive rankings. For 2006, it was a negative 0.08 correlation. For 2005, it was 0.32. For 2004, it was negative 0.11.

You are using the rankings rather than the DVOA values. They both show the same pattern -- the negative correlation is produced by the scaling of the DVOA values -- higher offensive is better whereas lower defensive is better.

For the 5 year period the correlation of the DVOA values (not the ranks) is -.17, indicating a consistent, albeit small, relationship between good offense and good defense over those years.

I can keep going, if you like, but I see very little overall correlation between having a good offense and having a good defense (measured by DVOA). What little correlation there is (which obviously doesn't equate to causation) probably can be explained by the fact that good teams try to load up for runs at the Super Bowl, while bad teams try to rebuild. So on average, teams tend to gravitate slightly more toward being good or bad on both sides of the ball, rather than good at one and bad at the other. But it's only slightly (as those correlations above seem to show).

The "little" correlation phrase doesn't strike me as relevant. Even a tiny correlation can have an important impact on game performance. The problem here seems that you are evaluating correlation against some standard value (e.g., must be .20 to be "small") rather than against the practical value of the relationship.

Further the correlation-causality issue should be clarified -- correlation does not mean causation - but it is a prerequisite for establishing cause. We can argue the chicken and egg point but as you know neither of us will establish cause with the data we have.

So why did our opponents throw deeper more often last season?

Because they were behind? Because the sacks that you mentioned put them in less favorable distance-down situations. Because 31 was in deep coverage? You tell me if you know why.

 

[AbeBeta]

You're unnecessarily adding complexity because you're looking for something else. You want a measure of the TEAM. We're looking for a measure of the DEFENSE.

I'm looking for a measure that associates most closely with the only outcome that has any relevance to the game.

 

[theogt]

I'm looking for a measure that associates most closely with the only outcome that has any relevance to the game.

Thats fine. As long as were both clear youre not looking for the most relevant measure of defense.

 

[ThreeSportStar80]

I actually hope Dallas signs Hamlin to a long term deal because he's a valued part of the secondary.

 

[AdamJT13]

This is what I mean. According to per-possession stats, the average score for 11 possessions should be up over 23 points, not 19.93.

And your sample was of well over 100 games.

You keep forgetting that the number of possessions within a game is affected by how each offense and defense is performing. The better the offenses are performing, generally, the fewer the possessions. The better the defenses are performing, generally, the more the possessions.

So, on average, defenses that faced 11 possessions were performing better than defenses that faced 10 possessions (they allowed fewer points and yards per possession). Yet they allowed more total points. Defenses that faced 11 possessions were performing better, on average, than defenses that faced nine possessions, but they also allowed more total points.

So, if they were performing better, why did they allow more total points? The only reason is because they faced more possessions, on average. That's why, as a measure of a defense, points per possession is better than points per game.

As a general rule, all other things being equal, it's not the number of possessions that determines how the defense performs (a defense doesn't perform better BECAUSE it faces 11 possessions instead of nine), it's how the defense performs that determines how many possessions it faces (a defense faces 11 possessions instead of nine BECAUSE it performs better than).

Your whole argument is based on the assumption that a team's points allowed per possession stays about the same as possessions are either added, or taken away. Remember?

No, my argument is that a defense's points allowed per possession stays the same if the performance level stays the same, regardless of how many possessions there are.

Your recent research does nothing to support this...

9 possessions
points allowed per possession: 2.11 (19.07 total)

10 possessions
your assumption: 2.11 (21.1 total)
actual number: 1.92 (19.2 total)
The jump up from 19.07 points to 19.2 is minoscule. That extra possession, based on your assumption, should have put the opposing offense up over 21 points. They ended up not even cracking 20!

Again, you're assuming that all of those defenses performed equally, which isn't the case. The defenses that faced more possessions were performing better, on average, which is why their points per possession is lower.

 

[AdamJT13]

It is negative -- which is why I wrote -.26 (note this is for DVOA not weighted DVOA which gives more weight to games later in the season, an adjustment that is irrelevant to the present discussion).

You are using the rankings rather than the DVOA values. They both show the same pattern -- the negative correlation is produced by the scaling of the DVOA values -- higher offensive is better whereas lower defensive is better.

For the 5 year period the correlation of the DVOA values (not the ranks) is -.17, indicating a consistent, albeit small, relationship between good offense and good defense over those years.

There is a .06 correlation between DVOA values (good offense to good defense) over the past five years and a 0.10 correlation between DVOA rankings. There's a negative .09 correlation between offensive and special teams DVOA rankings and a .02 correlation between defensive and special teams DVOA rankings.

All of those correlations are very small and probably can be explained either by chance or by the theory I already proposed -- that teams gravitate slightly toward being good or bad on both sides of the ball based on whether they're making a run at the Super Bowl (or playoffs) or simply rebuilding.

Because they were behind? Because the sacks that you mentioned put them in less favorable distance-down situations. Because 31 was in deep coverage? You tell me if you know why.

First of all, Roy was in deep coverage less often last year than the year before, yet our opponents attempted more deep passes. So that's not the reason.

Your other reasons are almost certainly correct -- they were behind more and in more long-yardage situations. so they had to throw deeper. Deeper passes naturally result in a higher YPA, which makes our drop in yards allowed per attempt even more signifcant.

 

[AbeBeta]

There is a .06 correlation between DVOA values (good offense to good defense) over the past five years and a 0.10 correlation between DVOA rankings. There's a negative .09 correlation between offensive and special teams DVOA rankings and a .02 correlation between defensive and special teams DVOA rankings.

All of those correlations are very small and probably can be explained either by chance or by the theory I already proposed -- that teams gravitate slightly toward being good or bad on both sides of the ball based on whether they're making a run at the Super Bowl (or playoffs) or simply rebuilding.

My data shows the correlation is -.14 between offense and defense. But regardless of whether it is .10 or .14, the idea that that variability can be explained both other factors based on your theory is not a reasonable claim. It is just as likely that the lack of relationship is explained by factors that play in the opposite direction and that if you take those factors into account the relationship increases. "Alternative explanations" cut both ways and do not exclusively decrease observed correlations.

To close the "small correlation" discussion, I would point you toward research on the effectiveness of aspirin in prevention of heart attacks. I'm sure you are aware that physicians recommend aspirin to reduce heart attacks - do you know the correlation between the taking aspirin and heart attack outcomes? r = .034. That's pretty small, huh. Practically that means however that among people who take aspirin you'll have about 2 or 3 fewer heart attacks compared to those who don't per 100.

Small correlations mean a ton, especially when focusing on important outcomes like heart attacks -- or performing well on the football field.

First of all, Roy was in deep coverage less often last year than the year before, yet our opponents attempted more deep passes. So that's not the reason.

Yanking your Roy chain never gets old.

 

[percyhoward]

My argument is that a defense's points allowed per possession stays the same if the performance level stays the same, regardless of how many possessions there are.

Right, and that a defense's points per game does not stay the same, even if the performance level stays the same, because it depends on how many possessions there are. Understood. The mistake there of course, is that it does not, and never did depend on how many possessions there are, which is why, when you expected to get numbers like this...

9 possessions 18 points
10 possessions 20 points
11 possessions 22 points
12 possessions 24 points
13 possessions 26 points

You instead got numbers like this...

9 possessions 19 points
10 possessions 19 points
11 possessions 20 points
12 possessions ???
13 possessions ???

(results that could just as easily have come up from comparing games played in cities that began with letters A-E, F-J, etc.)

And now to explain it you say that, yes, there is a correlation between number of possessions and points scored, but with the newfound caveat that a lot of possessions means better defensive play, and this could tend to counterract the fact that more possessions are giving the offense more chances to score.

All of which is a roundabout way of saying: more possessions don't result in more points.

And if more possessions don't result in more points, there's no reason to count possessions and factor it into a scoring stat.

If the offense's performance level is the same, scoring will go up as the number of possessions goes up. That's an inherent fact.

And yet, as possessions go up, defensive play goes up.

When the opponent had eight legitimate possessions, those defenses allowed an average of 18.08 points per game. With nine possessions, it was 19.07. With 10, it was 19.19. And with 11, it was 19.93. So yes, it appears that within a typical range of possessions, more possessions do mean more points, on average.

You showed data for games with 8-11 possessions, calling it a "typical range," but the average number of possessions in a game is actually a little over 11.

So a typical range would be 10-13, not 8-11.

What about 12-possession games?

Also, you broke down 10 teams' games. Did more possessions mean proportionally more points, team-by-team?

 

[AdamJT13]

My data shows the correlation is -.14 between offense and defense. But regardless of whether it is .10 or .14, the idea that that variability can be explained both other factors based on your theory is not a reasonable claim. It is just as likely that the lack of relationship is explained by factors that play in the opposite direction and that if you take those factors into account the relationship increases. "Alternative explanations" cut both ways and do not exclusively decrease observed correlations.

Maybe you can formulate a coherent counter-argument and post it.

To close the "small correlation" discussion, I would point you toward research on the effectiveness of aspirin in prevention of heart attacks. I'm sure you are aware that physicians recommend aspirin to reduce heart attacks - do you know the correlation between the taking aspirin and heart attack outcomes? r = .034. That's pretty small, huh. Practically that means however that among people who take aspirin you'll have about 2 or 3 fewer heart attacks compared to those who don't per 100.

If there's a consistent benefit -- no matter how small -- and no harmful side effects, then it's useful for medical purposes.

But, if multiple studies showed correlations that consistently varied between slightly positive (fewer heart attacks) and slightly negative (more heart attacks), I doubt doctors would consider that a worthwhile risk -- especially considering that the next study could swing the correlation to an overall negative, meaning that those taking aspirin have more heart attacks.

The correlation between offensive success and defensive success is small enough and varies enough -- not to mention the other causational factors -- that it's insignificant.

 

[AdamJT13]

You showed data for games with 8-11 possessions, calling it a "typical range," but the average number of possessions in a game is actually a little over 11. So a typical range would be 10-13, not 8-11.

Remember, I eliminated all possessions when the offense either didn't have time to score or had no intention of scoring. That eliminated an average of 1.2 possessions per team per game. The vast majority of games had 8-11 possessions for each team. With more or less than that, the sample size was smaller and the stats fluctuated wildly.

Right, and that a defense's points per game does not stay the same, even if the performance level stays the same, because it depends on how many possessions there are. Understood. The mistake there of course, is that it does not, and never did depend on how many possessions there are, which is why, when you expected to get numbers like this...

9 possessions 18 points
10 possessions 20 points
11 possessions 22 points
12 possessions 24 points
13 possessions 26 points

You instead got numbers like this...

9 possessions 19 points
10 possessions 19 points
11 possessions 20 points
12 possessions ???
13 possessions ???

(results that could just as easily have come up from comparing games played in cities that began with letters A-E, F-J, etc.)

And now to explain it you say that, yes, there is a correlation between number of possessions and points scored, but with the newfound caveat that a lot of possessions means better defensive play, and this could tend to counterract the fact that more possessions are giving the offense more chances to score.

All of which is a roundabout way of saying: more possessions don't result in more points.

They do, if the level of performance is the same.

In order for your "expected" point totals to happen, the number of possessions in a game would have to be unaffected by the performance of the offenses and defenses involved. But we know that's not true. Better defense and/or worse offense leads to more possessions, while worse defense and/or better offense leads to fewer possessions. That means worse offenses get more chances to make up for performing poorly, and better offenses get fewer chances. The result is that "flattening" of the curve (although it's still increasing).

If we look only at the offenses or defenses performing at the same level, you would see a curve much closer to what is expected, because their performance is about the same, and only the number of possessions changes.

For instance, the average offense in the 134 games I looked at (10 teams, 16 games each, minus 26 duplicates) gained 30.0 yards per possession. So let's look at all of the offenses that averaged 28.0 to 32.0 yards per possession and figure that they performed about the same. The average for those teams was 1.97 points per possession, so we would expect that to be true no matter how many possessions they had.

Here's the breakdown for those 47 offenses --

8 possessions (2 teams) = 10.0 ppg (1.25 ppp)
9 possessions (5 teams) = 17.2 ppg (1.91 ppp)
10 possessions (11 teams) = 19.91 ppg (1.99 ppp)
11 possessions (14 teams) = 23.36 ppg (2.12 ppp)
12 possessions (8 teams) = 23.75 ppg (1.98 ppp)
13 possessions (2 teams) = 28.0 ppg (2.15 ppp)
14 possessions (5 teams) = 24.6 ppg (1.76 ppp)

Given the small sample sizes, that's about as consistently close to the expected values as you could possibly expect.

And yet, as possessions go up, defensive play goes up.

Again, you're mistaking cause and effect. As defensive performance goes up, the number of possessions it faces typically goes up.

 

[AdamJT13]

Which of these offenses performed better, statistically?

Offense A's possessions
5 plays, 14 yards, PUNT
3 plays, 20 yards, TOUCHDOWN
3 plays, 8 yards, PUNT
3 plays, 15 yards (plus 13 in penalties), TOUCHDOWN
3 plays, 36 yards, TOUCHDOWN
3 plays, minus-4 yards, PUNT
6 plays, 50 yards, TOUCHDOWN
5 plays, 44 yards, TOUCHDOWN
3 plays, 6 yards, PUNT
7 plays, 52 yards, FIELD GOAL
3 plays, 4 yards, PUNT
3 plays, minus-1 yards, PUNT
6 plays, 14 yards, INTERCEPTION
6 plays, 25 yards, PUNT


Offense B's possessions
4 plays, 96 yards, TOUCHDOWN
13 plays, 74 yards, FIELD GOAL
5 plays, 65 yards, TOUCHDOWN
10 plays, 78 yards, TOUCHDOWN
10 plays, 69 yards, TOUCHDOWN
3 plays, minus-7 yards, PUNT
13 plays, 69 yards, FIELD GOAL