Team Efficiency - A rational discussion

Status
Not open for further replies.

AE 87

Helluva Engineer
Messages
13,015
In another thread, I posted my opinion that we have an "efficient" offense. In that thread, several people offered the assessment that our offense is "not efficient."

I'm interested in anyone who is willing to offer a rational argument in support of the claim that we are "not efficient": (1) a definition of "efficient" and why it's helpful; (2) a way of rationally assessing efficiency; and (3) a standard by which this efficiency can be labeled good or not good.

In support of the claim that we "are efficient," (1) I define "efficient" as points/drive (PPD). A team that scores 3o points in 15 drives is equally as productive as a team that scores 30 points in 10 drives; however, the second team is 1.5 times more efficient (3 ppd vs. 2 ppd). I consider this definition helpful because what matters in football is having more points than your opponents at the end of the game, not the season, and different games have different number of drives. If team 1 and team 2, from my example, play each other, they'll get the same number of drives, so defenses being equal team 2 will score 1.5 times more.

(2) PPD:
Points: 7*#rushing and passing TDs + 3* FGs made; This number normalizes the points in a way attributable to the offense.
Drives: The sum of rushing touchdowns, passing touchdowns, FG attempts, failed 4th down conversions, punts, lost fumbles and interceptions. [This calculation ignores drives ending in safeties (safeties often reflect starting-position problems which were not the fault of the offense) and may introduce false drives resulting from fumbled returns. However, when you consider the 140-180 possessions in a season, the error in drives is likely not significant.]
PPD=Points/Drives

(3) It seems to me that four main factors affect efficiency: scheme (including play calls), talent, execution, and opposition efficiency. Each one of these factors is difficult to quantify independently. (Although, an industrious person might try to quantify opposition efficiency through some opponents of opponents algorithm and use some sort of recruiting class ranking to quantify talent).

Of these four factors, three are attributes of the team: scheme, talent and execution. Still, the raw points/drive average would still be meaningful if you assume that opposition efficiency would largely average out across the season (for examination of one game, you'd have to look at opposition efficiency).

For example, by my calculations, the top 5 offenses for 2012 in PPD were Louisiana Tech, Texas A&M, Oregon, Alabama, and Kan St. Now, I think that when typical college football fans think about the top offenses of 2012, teams 2-4 on that list would probably get an "I can see that" response while many might still question how LTech got ranked #1 (though they did average 51.5 pts/game and were #1 in Scoring Offense, productivity). By comparison, footballoutsiders' efficiency rank (FEI) which does account for opposition and excludes garbage time, ranks these teams as 11, 2, 6, 5, and 21 respectively (apparently K State's opponents were not as good as LTech's on D). The FEI #1 for 2012, Baylor, was #6 in my raw PPD stat. So, I think that my raw PPD stat is a pretty decent indicator of efficiency in both Off ppd (OPPD) and Def ppd-allowed (DPPD).

Now, let's look at GT's OPPD (rank) and DPPD (rank) from 2009 to 2012 (note: rankings are based on more than 2 significant digits for convenience):
OPPD: 3.22 (3), 2.24 (57), 2.99 (14), 2.81 (23); Combined: 2.82 (11)
DPPD: 2.27 (86), 2.15 (66), 2.26 (80), 2.37 (77); Combined: 2.27 (83)

By looking at these numbers, it should be pretty clear that our offense has been pretty efficient while our defense has not been over the last four years.

Now let's consider some baseline numbers, #10 and #20 OPPD and DPPD, for 2009 - 2012.

OPPD #10, #20: 2.899, 2.649; 3.122, 2.802; 3.159, 2.732; 3.097, 2.915
DPPD #10, #20: 1.256, 1.477; 1.422, 1.552; 1.375, 1.520; 1.331, 1.500

From these numbers we might generalize:
Top 20 OPPD would be about 2.77 and Top 10 about 3.07
Top 20 DPPD would be about 1.51 and Top 10 about 1.35

So, I label our current offense as efficient or good because after 9 games, I feel we have enough data to get a feel for where we are.

GT 2013 (with three games left):
OPPD average is 2.89
DPPD average is 1.57.

From these numbers, it seems that we have a top 20 Offense in scoring efficiency and are sniffing close to a top 20 in Defense scoring efficiency. Of course, the trick is getting both teams to play to their potential for each game. (By way of comparison, GT's 2007 D, #21 in scoring D at 20.8 pts/game allowed, also had a DPPD of 1.57. Our 2007 OPPD was 2.01.)

I look forward to someone offering their argument for why we're not a good or efficient offense.
 

zyang31

Georgia Tech Fan
Messages
13
We're not an efficient offense because the vast majority of our points are scored against a select few opponents (who also happen to be bad). What you are doing is taking the 50+ points we scored against Syracuse and Elon, and applying them against Virginia Tech and BYU drives. You can not do that. You can not weigh each game the same.

No offense, because I see this happen a lot, but averages simply aren't meaningful in this context because the data is just not normally distributed. You're assuming offensive points per drive is consistently normally distributed across all games with similar parameters (mean of 2.89). It's not.Any conclusions derived from that assumption is just wrong.

If your assumption is true, that the OPPD is normally distributed with a 2.89 mean across all games, you're saying we're likely to score a similar amount of points against Elon as we would against Virginia Tech. LOL.
 

AE 87

Helluva Engineer
Messages
13,015
We're not an efficient offense because the vast majority of our points are scored against a select few opponents (who also happen to be bad). What you are doing is taking the 50+ points we scored against Syracuse and Elon, and applying them against Virginia Tech and BYU drives. You can not do that. You can not weigh each game the same.

No offense, because I see this happen a lot, but averages simply aren't meaningful in this context because the data is just not normally distributed. You're assuming offensive points per drive is consistently normally distributed across all games with similar parameters (mean of 2.89). It's not.Any conclusions derived from that assumption is just wrong.

If your assumption is true, that the OPPD is normally distributed with a 2.89 mean across all games, you're saying we're likely to score a similar amount of points against Elon as we would against Virginia Tech. LOL.

Thank you for your thoughtful response. I take no offense, because you've missed the point. Your last sentence could not be more wrong. The point of PPD is not to predict performance against an individual opponent. In fact, I calculate DPPD as well as OPPD. The point of PPD is to compare the relative offensive or defensive efficiency between different teams. I tried to make this clear by speaking of ranking teams by OPPD and DPPD. As I said, the assumption is that the spectrum of defensive strength of all your opponents is roughly comparable to the spectrum of defensive strength of the opponents of the team to which you are comparing your offense or defense. I supported that assumption by comparing raw ppd to the more sophisticated FEI.

Now, you have asserted that "we're not an efficient offense." I'm asking you to substantiate that claim with some rational basis: "I'm interested in anyone who is willing to offer a rational argument in support of the claim that we are "not efficient": (1) a definition of "efficient" and why it's helpful; (2) a way of rationally assessing efficiency; and (3) a standard by which this efficiency can be labeled good or not good." If you think I've abused the statistics, then show me how to use them correctly.

Please don't introduce "LOL" attitude into the conversation. I am sincerely interested in knowing if there is a rational basis to support the assertion that we're not an efficient offense. My suspicion is that people are simply reacting emotionally. As a result, I'm very frustrated by their posts. However, if they have a rational basis, I'd like to learn what it is.
 

daBuzz

Ramblin' Wreck
Messages
965
Thank you for your thoughtful response. I take no offense, because you've missed the point. Your last sentence could not be more wrong. The point of PPD is not to predict performance against an individual opponent. In fact, I calculate DPPD as well as OPPD. The point of PPD is to compare the relative offensive or defensive efficiency between different teams. I tried to make this clear by speaking of ranking teams by OPPD and DPPD. As I said, the assumption is that the spectrum of defensive strength of all your opponents is roughly comparable to the spectrum of defensive strength of the opponents of the team to which you are comparing your offense or defense. I supported that assumption by comparing raw ppd to the more sophisticated FEI.

Now, you have asserted that "we're not an efficient offense." I'm asking you to substantiate that claim with some rational basis: "I'm interested in anyone who is willing to offer a rational argument in support of the claim that we are "not efficient": (1) a definition of "efficient" and why it's helpful; (2) a way of rationally assessing efficiency; and (3) a standard by which this efficiency can be labeled good or not good." If you think I've abused the statistics, then show me how to use them correctly.

Please don't introduce "LOL" attitude into the conversation. I am sincerely interested in knowing if there is a rational basis to support the assertion that we're not an efficient offense. My suspicion is that people are simply reacting emotionally. As a result, I'm very frustrated by their posts. However, if they have a rational basis, I'd like to learn what it is.

AE 87,
For your stat to be meaningful to me, I would like to see it adjusted for a per team ranking. Like zyang31 suggested, I believe this offense vastly underperforms versus better defenses.

Maybe you could you take your PPD and use a weighted average of the opposing team's defensive ranking to adjust that value. In other words, 1.87PPD versus Virginia Tech with the #7 overall defense (I'm just throwing out a# here, I'm not actually taking the time to look it up) would be interesting versus a 3.01 PPD against the #147 ranked Elon.
 

AE 87

Helluva Engineer
Messages
13,015
AE 87,
For your stat to be meaningful to me, I would like to see it adjusted for a per team ranking. Like zyang31 suggested, I believe this offense vastly underperforms versus better defenses.

Maybe you could you take your PPD and use a weighted average of the opposing team's defensive ranking to adjust that value. In other words, 1.87PPD versus Virginia Tech with the #7 overall defense (I'm just throwing out a# here, I'm not actually taking the time to look it up) would be interesting versus a 3.01 PPD against the #147 ranked Elon.

Yeah, I actually talked about that improvement in my first post. However, to suggest that PPD is not meaningful until that is done is to say practically all stats are not meaningful since they aren't opponent adjusted. Also, FEI does the sort of thing you suggest, and I compared PPD to FEI.

I'm still interested in someone offering a justification for calling us bad or inefficient. If you don't like PPD, what stat do you use to rate our O to make thst claim?
 

IronJacket7

Helluva Engineer
Messages
2,544
In another thread, I posted my opinion that we have an "efficient" offense. In that thread, several people offered the assessment that our offense is "not efficient."

I'm interested in anyone who is willing to offer a rational argument in support of the claim that we are "not efficient": (1) a definition of "efficient" and why it's helpful; (2) a way of rationally assessing efficiency; and (3) a standard by which this efficiency can be labeled good or not good.

In support of the claim that we "are efficient," (1) I define "efficient" as points/drive (PPD). A team that scores 3o points in 15 drives is equally as productive as a team that scores 30 points in 10 drives; however, the second team is 1.5 times more efficient (3 ppd vs. 2 ppd). I consider this definition helpful because what matters in football is having more points than your opponents at the end of the game, not the season, and different games have different number of drives. If team 1 and team 2, from my example, play each other, they'll get the same number of drives, so defenses being equal team 2 will score 1.5 times more.

(2) PPD:
Points: 7*#rushing and passing TDs + 3* FGs made; This number normalizes the points in a way attributable to the offense.
Drives: The sum of rushing touchdowns, passing touchdowns, FG attempts, failed 4th down conversions, punts, lost fumbles and interceptions. [This calculation ignores drives ending in safeties (safeties often reflect starting-position problems which were not the fault of the offense) and may introduce false drives resulting from fumbled returns. However, when you consider the 140-180 possessions in a season, the error in drives is likely not significant.]
PPD=Points/Drives

(3) It seems to me that four main factors affect efficiency: scheme (including play calls), talent, execution, and opposition efficiency. Each one of these factors is difficult to quantify independently. (Although, an industrious person might try to quantify opposition efficiency through some opponents of opponents algorithm and use some sort of recruiting class ranking to quantify talent).

Of these four factors, three are attributes of the team: scheme, talent and execution. Still, the raw points/drive average would still be meaningful if you assume that opposition efficiency would largely average out across the season (for examination of one game, you'd have to look at opposition efficiency).

For example, by my calculations, the top 5 offenses for 2012 in PPD were Louisiana Tech, Texas A&M, Oregon, Alabama, and Kan St. Now, I think that when typical college football fans think about the top offenses of 2012, teams 2-4 on that list would probably get an "I can see that" response while many might still question how LTech got ranked #1 (though they did average 51.5 pts/game and were #1 in Scoring Offense, productivity). By comparison, footballoutsiders' efficiency rank (FEI) which does account for opposition and excludes garbage time, ranks these teams as 11, 2, 6, 5, and 21 respectively (apparently K State's opponents were not as good as LTech's on D). The FEI #1 for 2012, Baylor, was #6 in my raw PPD stat. So, I think that my raw PPD stat is a pretty decent indicator of efficiency in both Off ppd (OPPD) and Def ppd-allowed (DPPD).

Now, let's look at GT's OPPD (rank) and DPPD (rank) from 2009 to 2012 (note: rankings are based on more than 2 significant digits for convenience):
OPPD: 3.22 (3), 2.24 (57), 2.99 (14), 2.81 (23); Combined: 2.82 (11)
DPPD: 2.27 (86), 2.15 (66), 2.26 (80), 2.37 (77); Combined: 2.27 (83)

By looking at these numbers, it should be pretty clear that our offense has been pretty efficient while our defense has not been over the last four years.

Now let's consider some baseline numbers, #10 and #20 OPPD and DPPD, for 2009 - 2012.

OPPD #10, #20: 2.899, 2.649; 3.122, 2.802; 3.159, 2.732; 3.097, 2.915
DPPD #10, #20: 1.256, 1.477; 1.422, 1.552; 1.375, 1.520; 1.331, 1.500

From these numbers we might generalize:
Top 20 OPPD would be about 2.77 and Top 10 about 3.07
Top 20 DPPD would be about 1.51 and Top 10 about 1.35

So, I label our current offense as efficient or good because after 9 games, I feel we have enough data to get a feel for where we are.

GT 2013 (with three games left):
OPPD average is 2.89
DPPD average is 1.57.

From these numbers, it seems that we have a top 20 Offense in scoring efficiency and are sniffing close to a top 20 in Defense scoring efficiency. Of course, the trick is getting both teams to play to their potential for each game. (By way of comparison, GT's 2007 D, #21 in scoring D at 20.8 pts/game allowed, also had a DPPD of 1.57. Our 2007 OPPD was 2.01.)

I look forward to someone offering their argument for why we're not a good or efficient offense.

Lmbo! Oh dear Lord. Here we go again.
 

zyang31

Georgia Tech Fan
Messages
13
"As I said, the assumption is that the spectrum of defensive strength of all your opponents is roughly comparable to the spectrum of defensive strength of the opponents of the team to which you are comparing your offense or defense. I supported that assumption by comparing raw ppd to the more sophisticated FEI. "

Even if the spectrum of your defensive opponents are similar, you only have 6-7 data points to choose from right now. At the end of the season, you'll have 12-14, still not many. Your standard error is going to be huge. You can not meaningfully say that our PPD ranks favorably against other teams when the confidence intervals are so damn wide.

"
Now, you have asserted that "we're not an efficient offense." I'm asking you to substantiate that claim with some rational basis: "I'm interested in anyone who is willing to offer a rational argument in support of the claim that we are "not efficient": (1) a definition of "efficient" and why it's helpful; (2) a way of rationally assessing efficiency; and (3) a standard by which this efficiency can be labeled good or not good." If you think I've abused the statistics, then show me how to use them correctly."

I haven't done that, because you're the one making the claim that 'we're and efficient offense'. Therefore the burden of proof is on you. All the evidence you've submitted is based on flawed assumptions (all cupcakes are the same), limited pool of data (massive standard error)s, and an assumption that your PPD probability distribution per game will be the game (it's not).
 

AE 87

Helluva Engineer
Messages
13,015
"As I said, the assumption is that the spectrum of defensive strength of all your opponents is roughly comparable to the spectrum of defensive strength of the opponents of the team to which you are comparing your offense or defense. I supported that assumption by comparing raw ppd to the more sophisticated FEI. "

Even if the spectrum of your defensive opponents are similar, you only have 6-7 data points to choose from right now. At the end of the season, you'll have 12-14, still not many. Your standard error is going to be huge. You can not meaningfully say that our PPD ranks favorably against other teams when the confidence intervals are so damn wide.

"
Now, you have asserted that "we're not an efficient offense." I'm asking you to substantiate that claim with some rational basis: "I'm interested in anyone who is willing to offer a rational argument in support of the claim that we are "not efficient": (1) a definition of "efficient" and why it's helpful; (2) a way of rationally assessing efficiency; and (3) a standard by which this efficiency can be labeled good or not good." If you think I've abused the statistics, then show me how to use them correctly."

I haven't done that, because you're the one making the claim that 'we're and efficient offense'. Therefore the burden of proof is on you. All the evidence you've submitted is based on flawed assumptions (all cupcakes are the same), limited pool of data (massive standard error)s, and an assumption that your PPD probability distribution per game will be the game (it's not).

You have also made a positive claim: that our team is not efficient. I'm wanting to know if there is anybody willing to back that up with any kind of testable evidence.
 

zyang31

Georgia Tech Fan
Messages
13
You have also made a positive claim: that our team is not efficient. I'm wanting to know if there is anybody willing to back that up with any kind of testable evidence.
Fine dude, if you're going to be nitpicking my wording like that, I ammend my original post to

"Your argument shows several flaws and shortcomings due to (reasons I have listed above). I therefore do not think there is sufficient evidence to accept your hypoethesis that are offensive is highly efficient relative to other teams."

Happy now?
 

AE 87

Helluva Engineer
Messages
13,015
Fine dude, if you're going to be nitpicking my wording like that, I ammend my original post to

"Your argument shows several flaws and shortcomings due to (reasons I have listed above). I therefore do not think there is sufficient evidence to accept your hypoethesis that are offensive is highly efficient relative to other teams."

Happy now?

No. You did not understand my post, which I have pointed out. You've also missed the point of the thread. I'm not as much trying to persuade you that we have an efficient offense as wanting to know why you and others don't think we do. There's no "We suck until proven good" rule. If you think we're not efficient, why do you think that?
 

zyang31

Georgia Tech Fan
Messages
13
AE87, here's a quick sanity test to you. What's more efficient? Scores of 50 and 30 against Elon and VT respectively, or scores of 70 and 10 against them respectively? Assuming similar defensive output from us. I don't see much difference between 50 and 70, but see a helluva difference between 30 and 10 in practice.

If your stats don't match what's in reality, your stats are probably wrong. That brings me to another thought, even if we accept the assumption that all teams face a similar spectrum of defensive strength, and that spectrum does not introduce too much noise into the data, I still don't think we can weight each point the same.

There is no effective difference between 70 points and 50 points in a blowout against a cupcake, but there is a HUGE difference between 30 points and 10 points in a close game. But in your analysis, the difference between 70 and 50 vs Elon is the same as the difference between 30 and 10 vs VT. That discrepancy is not accounted for in your analysis.

"There's no "We suck until proven good" rule. If you think we're not efficient, why do you think that?"

Stop putting words in my mouth. I did not say we suck. I'm saying there is no evidence to suggest GT's offensive efficiency is significantly better than most other teams.

Take a course in statistics. The default hypothesis is there is no significant difference between variables unless shown otherwise. It'

I understand exactly what you're saying.

If we assume every team faces a similar defensive spectrum and
If every team has similar offensive output against defensive strengths (IE every team scores around 70 points against their cupcakes and around 10 points against top defenses like us (40 points a game), and not say only 40 against their cupcakes but 28 against the top defense(34 points a game)) and
If the extremely limited data sample does not intoduce result in massive standard errors

then you're right, GT is an efficient team.

But you'll have to pardon me if I don't think treating point scored against Elon as if it were scored against VT is good statistical analysis.
 
Last edited:

IronJacket7

Helluva Engineer
Messages
2,544
You have also made a positive claim: that our team is not efficient. I'm wanting to know if there is anybody willing to back that up with any kind of testable evidence.

We may beat Clemson and UGA. I hope we turn the corner and do so...

But to respond one time on this "rational thread", the only "testable" evidence I can think of is that when we have played good teams (miami, byu, vatech) we have lost and our offense have not been efficient (turnovers, 3 and outs, penalties)

Is that "TESTABLE" enough for you Dr.?

Sorry... I didn't include any DDP, PPD, FEI, WCW, NWO, ABCD stats in there...
 

AE 87

Helluva Engineer
Messages
13,015
We may beat Clemson and UGA. I hope we turn the corner and do so...

But to respond one time on this "rational thread", the only "testable" evidence I can think of is that when we have played good teams (miami, byu, vatech) we have lost and our offense have not been efficient (turnovers, 3 and outs, penalties)

Is that "TESTABLE" enough for you Dr.?

Sorry... I didn't include any DDP, PPD, FEI, WCW, NWO, ABCD stats in there...

No. This answer provides no definition of efficient and no way of comparing GT's offense to other offenses. Of course, we didn't perform as well against good defenses as we did against bad defenses. Please, don't try to spam this thread with your sarcasm. If you don't have a rational response to the questions, don't post.
 

zyang31

Georgia Tech Fan
Messages
13
Answer the question AE87, which is more efficient. 70 points vs Elon and 10 points vs VT or 40 points vs Elon and 28 points vs VT assuming the same defensive output?
 
Status
Not open for further replies.
Top