My vote for NFL MVP through six weeks (1 Viewer)

[moleculo]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

 

[twitch]

does anyone think Garrard would have 21 and 2 with an entire All-Pro team, let alone just one All-Pro (Moss). No. 21TDs. 2INTs. But the real reason Brady is the MVP not because of those stats. Its because he's brought Randy Moss back from the dead. No other QB this year has done that.

 

[cobalt_27]

This appears to be the classic case where statistical analyses lead one to completely lose his mind and make the silliest inferences.

Chase, may I suggest you just slink away from this one. You'll thank me later, trust me.

 

[831hoops]

I think you already demonstrated he usefulness of this stat when you used it support your argument that the Raiders would be a top 5 defense this year.

 

[cobalt_27]

Numbers lie.

The numbers are what the numbers are. They don't tell lies.Our extrapolations and faulty inferences based on the numbers lie.

 

[cobalt_27]

I think you already demonstrated he usefulness of this stat when you used it support your argument that the Raiders would be a top 5 defense this year.

Oh, those were the good ol' days.

 

[moleculo]

Numbers lie.

The numbers are what the numbers are. They don't tell lies.Our extrapolations and faulty inferences based on the numbers lie.

 

[moleculo]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

 

[cobalt_27]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

 

[pollardsvision]

not sure if i'd go with MVP, but i've watched every snap and he's played beautifully. he's carried the offense during the times that the running game has stalled.

he's been nearly flawless and i can't really recall any bad throws on his part (not to say that their haven't been).

when the offense stalls, it's generally been the running game or dropped passes (or reggie williams fumbles).

as a jags fan, i'm pretty fired up about garrard. i just hope we're having this conversation 2.5 months from now.

 

[Tony Jabroni]

I 'gave up' on Chase Stuart when he went to the #### bar "for the food".

True story.

 

[Chase Stuart]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

I have no idea why you think this. It's inaccurate. How Brady does has absolutely nothing to do with a nuanced strength of schedule adjustment. I'm not sure why this is difficult to comprehend.

 

[Chase Stuart]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

This is inaccurate.I don't understand how you'd look at six weeks worth of data and not do SOS adjustment. It's significantly more important to do after six weeks than after 16 weeks (do you understand why?). The way the SOS adjustment has been done here, Brady and Garrard's performance does not contribute to their opponent's SOS rankings.

 

[Chase Stuart]

does anyone think Garrard would have 21 and 2 with an entire All-Pro team, let alone just one All-Pro (Moss). No. 21TDs. 2INTs. But the real reason Brady is the MVP not because of those stats. Its because he's brought Randy Moss back from the dead. No other QB this year has done that.

Brady is a much better QB than Garrard. In all likelihood, Brady's stats will dwarf Garrard's stats for the rest of the season.

 

[moleculo]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

I have no idea why you think this. It's inaccurate. How Brady does has absolutely nothing to do with a nuanced strength of schedule adjustment. I'm not sure why this is difficult to comprehend.

I'm not sure why this is difficult to comprehend either. Let's try an experiment. What do your AYPA look like after week 1? Without doing the math, I can tell you that the SOS adjusted rankings would have everyone being exactly at the average. Let's say Bradys ARPA was 9 hypothetically, and the league average was 6. Brady was therefore at +3. Conversely, the Jet's defensive AYPA would be a -3. Add them together, and you get zero.

So, we know that SOS is complete garbage, one week in.

Now, do the rankings after two weeks. Let's assume for the sake of discussion that Brady put up 9 again, and the league average remained 6. Let's also assume that the Jet's gave up the league average, 6, to Baltimore, and SD in week 1 gave up 6 to Chicago. Now, the Jets and Chargers SOS score would be -1.5, the average of -3 and 0.

So, at this point, Brady has scored 3 AYPA above the league average in two consecutive games in this hypothetical exercise. However, the defenses he has faced are averaging -1.5 AYPA below the league average. So, when you adjust Brady's score for SOS, you would get 1.5 AYPA.

In both week 1 and week 2, when we applied the SOS to Brady's score, it dropped because he has faced inferior defenses. However, are those defenses really that bad? In games that they played against non-Brady teams, they turned in average scores, but their scores are lower only due to the performances by Brady. Therefore, Brady himself has effected the SOS scores of his opponents directly, and so when you apply SOS to Brady's numbers, you are actually giving him a reduction.

The effect of this recursive is reduction reduced with every game - after one week, Brady accounts for 100% of what his opponents have faced and so his SOS is 100% dependant on what he did. After week 2, he accounts for 50%. After week 3, 33%, and so on. The more games played, the better off this model is. How many games is enough for this to be valid? i have no idea, but I'm pretty sure it's >6, and I would guess it is closer to 30.

 

[moleculo]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

This is inaccurate.I don't understand how you'd look at six weeks worth of data and not do SOS adjustment. It's significantly more important to do after six weeks than after 16 weeks (do you understand why?). The way the SOS adjustment has been done here, Brady and Garrard's performance does not contribute to their opponent's SOS rankings.

can you unpack that a little? How can it not?

 

[Doug Drinen]

Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that.

It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

There is a way around the chicken-egg problem. It's called iteration. It's pretty darn cool. Let me walk you through it and try to explain why it does indeed factor out the situation you're concerned about.STEP 1: compute an unadjusted rating for every QB and for every defense.

At this point, Brady's rating is too high (because it hasn't been adjusted to account for the weak defenses he's played). Meanwhile, assuming they've faced a more-or-less average collection of QBs aside from Brady, the Bills, Jets, Cowboys, etc have defensive ratings that are too low (because they haven't been adjusted to account for the strong QBs (i.e. Brady) that they've faced).

STEP 2: adjust every QB's rating based on the average defense he has faced.

At this point, Brady's rating is likely closer to being "correct," but is probably too low because we've used the STEP 1 defense ratings, which (probably) underrate the defenses Brady has faced. This is your main point: that we have overcorrected Brady's numbers. And if we stopped here, we would indeed have overcorrected (probably).

STEP 3: now adjust each defense's rating based on the average QB it has faced.

Again assuming the Patriots' opponents have faced a fairly average collection of QBs, except for Brady, this will improve the rating of each of those defenses by the full amount by which Brady is better than average.

At this point, it's pretty subtle trying to figure out whether Brady and/or his opponents have been over- or under-corrected, but the next steps will take care of it.

STEPS 4 through 1000: repeat steps 2 and 3 over and over again. Adjust Brady again based on the new defensive estimates, and adjust the defenses again based on the new QB estimates. Eventually, the numbers will stop changing.

Let me show you how it works specifically in this case:

Brady's nominal AYPA is +3.298 yards per attempt above average. That's what the raw stat line says.

Nominally, the defenses he has faced look like this:

Jets: -1.61 (<---- 1.61 yards per pass WORSE than average)

Chargers: -.57

Bills: -.00

Bengals: -.54

Browns: -.73

Cowboys: +.73

Now, if you take a weighted average of those six numbers, weighted by how many attempts Brady had against each team, you get -.36. So in the first pass, we'd decrease Brady's rating from +3.298 to about +2.94. [NOTE: if we stopped here, your criticism would be dead on. That's why we're not stopping here ] Meanwhile, the defenses that faced Brady get a collective boost. Here's where they stand after one step:

Jets: -1.48

Chargers: -.40

Bills: +.97

Bengals: +.28

Browns: -.55

Cowboys: +.31

Note how all the defenses that played against Brady are better than they appear (except the Cowboys, who have faced Bulger, Edwards, Green, and Grossman). This is your point moleculo, and it's a good point, which is why we're not finished yet.

So after we re-adjust Brady based on those revised defensive numbers, Brady moves back up from +2.94 to +3.18. Then we just do the whole thing over and over and over again. As the process moves on, Brady's ratings do the following:

3.30 ---> 2.94 ---> 3.18 ---> 3.04 ---> 3.14 ---> 3.07 ---> 3.12 ---> 3.09 ---> 3.12 ---> 3.11 ---> 3.12 ---> 3.13 ---> 3.13 ---> 3.13 ---> ...

And here are where the Pats' opponent's numbers eventually settled in:

Jets: -1.26

Chargers: -.27

Bills: +1.19

Bengals: -.21

Browns: -1.20

Cowboys: +.60

Note that each of those defenses has been adjusted by the full amount that the QBs they've faced are better or worse than average. For example, look at the Bills.

The Bills have faced:

Brady (+3.13)

Romo (+1.63)

Cutler (+.84)

Big Ben (+.54)

Pennington (+.12)

The weighted average of those is +1.20, which is exactly the amount that the iteration moved the Bills' rating. Remember, the stat line said the Bills are exactly average (-.00). After iteration, we find that Bills' schedule has been 1.2 better than average, so they end up with a rating of 1.2, which is the number we're ultimately using to adjust Brady.

Here is another way to see it:

The raw stats said Brady's schedule was .36 worse than average. After iteration, we figure that it's .16 worse than average. So the iteration bumped these teams up by a collective .20. Let's examine where that came from. For starters, Brady is 3.13 better than average. Now, if you go look at the *other* QBs faced by teams faced by Brady, you'll find that they are, on (weighted) average, about .38 worse than average (I can list them and their ratings if you want). Since about 1/6 of the schedule was against Brady, he bumped their ratings by about 3.13*(1/6). Meanwhile, the other QBs, which account for 5/6 of their collective schedule, bumped them down by about .38*(5/6). Since 3.13*(1/6) - .38*(5/6) is just about .2, that's why Brady's schedule got a boost of .2 for playing against Brady, along with a collection of other QBs that were a little below average.

It's all extremely self-referential (the chicken-egg thing your referred to earlier). It is, in a sense, infinitely self-referential. But it all balances. I think it's pretty cool.

EDIT: Just to pre-empt a potential hijack....

I used AYPA in the above example, but all of the above is agnostic about whether AYPA is worth anything or not. [For the record, I happen to like it, but I don't have as much energy as young Chase, so I'll let him answer those things]. This post is only about the SOS adjustment. You could do the same with (almost) any stat.

 

[moleculo]

Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that.

It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

There is a way around the chicken-egg problem. It's called iteration. It's pretty darn cool. Let me walk you through it and try to explain why it does indeed factor out the situation you're concerned about.STEP 1: compute an unadjusted rating for every QB and for every defense.

At this point, Brady's rating is too high (because it hasn't been adjusted to account for the weak defenses he's played). Meanwhile, assuming they've faced a more-or-less average collection of QBs aside from Brady, the Bills, Jets, Cowboys, etc have defensive ratings that are too low (because they haven't been adjusted to account for the strong QBs (i.e. Brady) that they've faced).

STEP 2: adjust every QB's rating based on the average defense he has faced.

At this point, Brady's rating is likely closer to being "correct," but is probably too low because we've used the STEP 1 defense ratings, which (probably) underrate the defenses Brady has faced. This is your main point: that we have overcorrected Brady's numbers. And if we stopped here, we would indeed have overcorrected (probably).

STEP 3: now adjust each defense's rating based on the average QB it has faced.

Again assuming the Patriots' opponents have faced a fairly average collection of QBs, except for Brady, this will improve the rating of each of those defenses by the full amount by which Brady is better than average.

At this point, it's pretty subtle trying to figure out whether Brady and/or his opponents have been over- or under-corrected, but the next steps will take care of it.

STEPS 4 through 1000: repeat steps 2 and 3 over and over again. Adjust Brady again based on the new defensive estimates, and adjust the defenses again based on the new QB estimates. Eventually, the numbers will stop changing.

Let me show you how it works specifically in this case:

Brady's nominal AYPA is +3.298 yards per attempt above average. That's what the raw stat line says.

Nominally, the defenses he has faced look like this:

Jets: -1.61 (<---- 1.61 yards per pass WORSE than average)

Chargers: -.57

Bills: -.00

Bengals: -.54

Browns: -.73

Cowboys: +.73

Now, if you take a weighted average of those six numbers, weighted by how many attempts Brady had against each team, you get -.36. So in the first pass, we'd decrease Brady's rating from +3.298 to about +2.94. [NOTE: if we stopped here, your criticism would be dead on. That's why we're not stopping here ] Meanwhile, the defenses that faced Brady get a collective boost. Here's where they stand after one step:

Jets: -1.48

Chargers: -.40

Bills: +.97

Bengals: +.28

Browns: -.55

Cowboys: +.31

Note how all the defenses that played against Brady are better than they appear (except the Cowboys, who have faced Bulger, Edwards, Green, and Grossman). This is your point moleculo, and it's a good point, which is why we're not finished yet.

So after we re-adjust Brady based on those revised defensive numbers, Brady moves back up from +2.94 to +3.18. Then we just do the whole thing over and over and over again. As the process moves on, Brady's ratings do the following:

3.30 ---> 2.94 ---> 3.18 ---> 3.04 ---> 3.14 ---> 3.07 ---> 3.12 ---> 3.09 ---> 3.12 ---> 3.11 ---> 3.12 ---> 3.13 ---> 3.13 ---> 3.13 ---> ...

And here are where the Pats' opponent's numbers eventually settled in:

Jets: -1.26

Chargers: -.27

Bills: +1.19

Bengals: -.21

Browns: -1.20

Cowboys: +.60

Note that each of those defenses has been adjusted by the full amount that the QBs they've faced are better or worse than average. For example, look at the Bills.

The Bills have faced:

Brady (+3.13)

Romo (+1.63)

Cutler (+.84)

Big Ben (+.54)

Pennington (+.12)

The weighted average of those is +1.20, which is exactly the amount that the iteration moved the Bills' rating. Remember, the stat line said the Bills are exactly average (-.00). After iteration, we find that Bills' schedule has been 1.2 better than average, so they end up with a rating of 1.2, which is the number we're ultimately using to adjust Brady.

Here is another way to see it:

The raw stats said Brady's schedule was .36 worse than average. After iteration, we figure that it's .16 worse than average. So the iteration bumped these teams up by a collective .20. Let's examine where that came from. For starters, Brady is 3.13 better than average. Now, if you go look at the *other* QBs faced by teams faced by Brady, you'll find that they are, on (weighted) average, about .38 worse than average (I can list them and their ratings if you want). Since about 1/6 of the schedule was against Brady, he bumped their ratings by about 3.13*(1/6). Meanwhile, the other QBs, which account for 5/6 of their collective schedule, bumped them down by about .38*(5/6). Since 3.13*(1/6) - .38*(5/6) is just about .2, that's why Brady's schedule got a boost of .2 for playing against Brady, along with a collection of other QBs that were a little below average.

It's all extremely self-referential (the chicken-egg thing your referred to earlier). It is, in a sense, infinitely self-referential. But it all balances. I think it's pretty cool.

Doug,With all due respect, I still don't buy into this. I understand the math - I do have a background including optimization (from an engineers perspective), so I understand where you are coming from. The fact that it does converge is pretty cool.

However, I still have significant issues with this procedure. I am about to have 20-30 houseguests show up any minute now for my annual Octoberfest party so I can't get into it; I should have my rebuttal tomorrow.

 

[cobalt_27]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

This is inaccurate.I don't understand how you'd look at six weeks worth of data and not do SOS adjustment. It's significantly more important to do after six weeks than after 16 weeks (do you understand why?). The way the SOS adjustment has been done here, Brady and Garrard's performance does not contribute to their opponent's SOS rankings.

Before I respond, would indulge me in just how you are computing SOS here and provide some actual raw data for the JAX and NE opponents. There are two issues at play here, but I am making some assumptions on how you're calculating the data. I just want to be sure I understand what you're doing with those data before offering a response.

 

[Chase Stuart]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

This is inaccurate.I don't understand how you'd look at six weeks worth of data and not do SOS adjustment. It's significantly more important to do after six weeks than after 16 weeks (do you understand why?). The way the SOS adjustment has been done here, Brady and Garrard's performance does not contribute to their opponent's SOS rankings.

Before I respond, would indulge me in just how you are computing SOS here and provide some actual raw data for the JAX and NE opponents. There are two issues at play here, but I am making some assumptions on how you're calculating the data. I just want to be sure I understand what you're doing with those data before offering a response.

Please see Doug's post. He did a terrific job explaining it.

 

[NE_REVIVAL]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

Chase isn't looking outside the box; he is creating a box that shockingly just happens to fit a well documented bias. You need look no further than the irrational nonsense he wrote earlier in the year with regard to Bradys career and SB seasons. No offense to Doug or other staff members here at FBG, but the paralysis by analysis and attempts to use statistics and faulty subjective logic to ignore what common sense and anyone with 2 eyes can see is comical. There are complex elements in football that are virtually impossible to account for in any meaningful. And there are other important statistics such as Wins & losses that are often left out all togethor.

You don't need a statistical formula to see that at this point in the season Manning & Brady are head & shoulders better candidates for MVP than Garrard. If the formula you are using indicates otherwise then you need to consider the obvious possibility that your formula is seriously flawed. Maybe you are putting too much emphasis on certain stats and not enough on others? Are you accounting for points scored(all)? First downs? Red Zone offense, 2 minute offense? How about playing from behind? Wins & Losses? Etc etc.

Finally, since we are talking about the MVP and we agree that wins are more important than OVA How about we simply switch qbs? We can't say for sure that JAX would be undefeated with Manning or Brady at qb, but I think most would agree it is much more likely than not. Can anyone make a serious argument that it is likely NE or INDY would still be undefeated with Garrard at qb?

 

[Tau837]

Still wondering why 10 yards per TD pass is right. I see a basis for the 45 yards per interception. Did I miss the explanation on the TD pass value?

 

[Workhorse]

What interesting to me is that Garrard is a scrambling QB who has been sacked 15 times so far this season.

Tom Brady is a classic pocket QB who has only been sacked 6 times.

Is the New England O-Line that much better than the Jacksonville line?

 

[Chase Stuart]

What interesting to me is that Garrard is a scrambling QB who has been sacked 15 times so far this season.Tom Brady is a classic pocket QB who has only been sacked 6 times.Is the New England O-Line that much better than the Jacksonville line?

Scrambling QBs almost always get sacked more often than pocket QBs, believe it or not. But yes, I think the Pats OL is the best pass blocking unit in the league. I'm not even sure which team comes close, outside of Indy's. The Pats' OL, before rightly getting its due this season, was the most underrated unit in the entire NFL IMO.

 

[NE_REVIVAL]

What interesting to me is that Garrard is a scrambling QB who has been sacked 15 times so far this season.Tom Brady is a classic pocket QB who has only been sacked 6 times.Is the New England O-Line that much better than the Jacksonville line?

Great point, thats a HUGE one that I forgot to list. A large part of what makes Manning and Brady so good is their ability to anticiapte the defense and get rid of the ball before they get sacked. I don't know that there is anyone else in the league even close. I doubt this innate abilty is factored in to the OVA formula, but I could be wrong.

 

[Chase Stuart]

I've given lots of statistical arguments for why Garrard has been great this year, but I'll write something here to appease the other crowd. FWIW, it's important to remember that I don't think Garrard is anywhere near as good as Brady or Manning, and he's not going to play anywhere near as well as Brady or Manning the rest of the way. What he has done, however, is play just as well as them so far this year. (If we were to start the season over again, I don't think Garrard would play this well again, and Jacksonville would probably be 3-2 or 2-3).

Week 1: The Jaguars allow 246 rushing yards to Tennessee, while Jaguars RBs rush for 48 yards. Up 10-6 in the second half, Jacksonville can't try a 36 yard field goal due to an injury to their kicker. Late in the game, with Jacksonville down three, Marcedes Lewis runs into MJD, and Drew fumbles inside the TEN 10 yard line. Garrard throws a 47 yard TD to John Broussard, and throws for 200 yards and no interceptions against a Titans pass defense that made Peyton Manning and Drew Brees have their worst games of the season. Tennessee so far this season, has been a top three pass defense.

Week 2: Against Atlanta, Garrard throws for 272 yards on 25 pass attempts. Garrard was 7-of-8 for 122 yards on the two fourth-quarter scoring drives. He also completes four passes of over 20 yards.

Week 3: Against Denver, a team that had been playing by far the best pass defense in the NFL so far, Garrard completes 70% of his passes in an efficient performance with 1 TD and 0 INT. He also runs for 52 yards. On 3rd and 4 from the Denver 20 late in the game, Garrard runs for 19 yards.

Week 5: Against Kansas City, Garrard averages over 8 yards per pass, chipping in a TD and 0 INT. The Jaguars beat Kansas City in Arrowhead, where they've won 14 of their last 16 games outside of this one. Garrard ices the game with a 40 yard pass in the 4th quarter, followed by his TD throw. This Chiefs pass D would intercept Carson Palmer twice, and 7.4 yards per pass, the following week.

Week 6: Garrard throws for over 200 yards and 2 scores, beating Houston. He also runs for 31 yards. Jacksonville finally beats Houston, a team they've struggled against in the past.

Meanwhile, look at who Garrard is doing this with: Reggie Williams, Dennis Northcutt, MJD and Marcedes Lewis are the only receivers with 100 yards on the year. Jacksonville's receivers are obviously way worse than Indianapolis', Dallas' or New England's wide receivers, but Jacksonville is still 4-1. The only game they lost was when their defense got shredded, their kicker got hurt, and they had a bad fumble late in the game. Garrard has had a QB Rating over 100 in four consecutive games.

Garrard isn't that good. But he's been playing out of his mind good.

 

[zoonation]

Imagine how much better the Patriots would look so far if they had Garrard

The MVP award isn't given to the best player, it's given to the player who has played the best. Obviously Brady is a way better QB than Garrard.Exactly what has Garrard done wrong this year?

this is not the criteria for MVP of the NFL

 

[pollardsvision]

this has nothing to do with the fancy numbers, but when i watch david garrard in his 1st 5 games this season, he looks to me to be what vince young should aspire to be.

or rather, garrard is playing like a less-explosive version of VY could be one day.

decisive, accurate, and makes plays with his feet when it's necessary. there's not big passing #s b/c the team is still oriented around a ball-control, running game and defense. the QB needs to efficient , accurate and make drive sustaining runs that keep the defense off balance.

this may sound like a "game manager", but it's not. garrard is winning games for the jags.

 

[cobalt_27]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

This is inaccurate.I don't understand how you'd look at six weeks worth of data and not do SOS adjustment. It's significantly more important to do after six weeks than after 16 weeks (do you understand why?). The way the SOS adjustment has been done here, Brady and Garrard's performance does not contribute to their opponent's SOS rankings.

Before I respond, would indulge me in just how you are computing SOS here and provide some actual raw data for the JAX and NE opponents. There are two issues at play here, but I am making some assumptions on how you're calculating the data. I just want to be sure I understand what you're doing with those data before offering a response.

Please see Doug's post. He did a terrific job explaining it.

Ok, I missed that earlier. I like the adjustment.Still, I think the inference is off. I'm thinking of a scenario here whereby you could have a QB be flawless and still rank lower than Garrard. And, when I say flawless, I mean flawless. This hypothetical QB could play all 16 games, complete 100% of his pass attempts, throw 6 TDs every game, throw for 7680 yards in a season and have zero interceptions and still rank lower than Garrard on your MVP list (and probably lower than David Carr, for that matter).

This hypothetical QB could line up at center and drop back to pass 120 times per game and complete 4-yarders on every play. First down: 4yd completion; second and 6 to go. Second down: 4yd completion; third and 2 to go. Third down: 4yd completion; 1st and 10. This QB has such uncanny accuracy that he never misses. He drives his team down the field on every possession and scores every time. It's that automatic. At the end of every game, he accounts for 6 TD passes and 480 yards passing on his 120 attempts.

If I understand your bias towards the AY/A, he would finish the season with a mark of 4.5 {[(7680 PaYds + (96TDs*10)) / 1920 attempts] = 8640/1920 = 4.5 AY/A}. This would be -1.47 below the league average (5.97). No matter what adjustments you made for SOS, he would still be ranked lower on your assessment than virtually all QBs in the game.

And, yet, in my mind, this hypothetical QB would be so off-the-charts in terms of his accuracy and the contributions he brings to his team that he would be the slam dunk of all slam dunks for MVP.

Brady isn't that QB. But, what he is doing is in a different realm than what Garrard is doing, who by his own right, is contributing so much to his team's success thus far. While I think statistical analyses can/should guide interpretation(s), they miss important variables along the way. We do what we can to reduce those misses, but in large part, I think your current MVP analysis misses some important qualitative features that Brady is contributing above/beyond Garrard.

Again, I do like your analysis, and it gives us a peek into their respective contributions (adjusting for the level of competition, to boot). But, at the end of the day, I don't think the conclusion you've reached is accurate.

Brady is the clear-cut MVP, thus far.

 

[Chase Stuart]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

This is inaccurate.I don't understand how you'd look at six weeks worth of data and not do SOS adjustment. It's significantly more important to do after six weeks than after 16 weeks (do you understand why?). The way the SOS adjustment has been done here, Brady and Garrard's performance does not contribute to their opponent's SOS rankings.

Before I respond, would indulge me in just how you are computing SOS here and provide some actual raw data for the JAX and NE opponents. There are two issues at play here, but I am making some assumptions on how you're calculating the data. I just want to be sure I understand what you're doing with those data before offering a response.

Please see Doug's post. He did a terrific job explaining it.

Ok, I missed that earlier. I like the adjustment.Still, I think the inference is off. I'm thinking of a scenario here whereby you could have a QB be flawless and still rank lower than Garrard. And, when I say flawless, I mean flawless. This hypothetical QB could play all 16 games, complete 100% of his pass attempts, throw 6 TDs every game, throw for 7680 yards in a season and have zero interceptions and still rank lower than Garrard on your MVP list (and probably lower than David Carr, for that matter).

This hypothetical QB could line up at center and drop back to pass 120 times per game and complete 4-yarders on every play. First down: 4yd completion; second and 6 to go. Second down: 4yd completion; third and 2 to go. Third down: 4yd completion; 1st and 10. This QB has such uncanny accuracy that he never misses. He drives his team down the field on every possession and scores every time. It's that automatic. At the end of every game, he accounts for 6 TD passes and 480 yards passing on his 120 attempts.

If I understand your bias towards the AY/A, he would finish the season with a mark of 4.5 {[(7680 PaYds + (96TDs*10)) / 1920 attempts] = 8640/1920 = 4.5 AY/A}. This would be -1.47 below the league average (5.97). No matter what adjustments you made for SOS, he would still be ranked lower on your assessment than virtually all QBs in the game.

And, yet, in my mind, this hypothetical QB would be so off-the-charts in terms of his accuracy and the contributions he brings to his team that he would be the slam dunk of all slam dunks for MVP.

Brady isn't that QB. But, what he is doing is in a different realm than what Garrard is, who by his own right, is contributing so much to his team's success thus far. While I think statistical analyses can/should guide interpretation(s), they miss important variables along the way. We do what we can to reduce those misses, but in large part, I think your current MVP analysis misses some important qualitative features that Brady is contributing above/beyond Garrard.

Again, I do like your analysis, and it gives us a peek into their respective contributions (adjusting for the level of competition, to boot). But, at the end of the day, I don't think the conclusion you've reached is accurate.

Brady is the clear-cut MVP, thus far.

Brady's AY/A is higher than any QB has ever had since the merger. AY/A boosts Brady, it doesn't hurt him. His AY/A is slightly higher than Manning's was in 2004.

 

[cobalt_27]

My problem with this is that Brady has accounted for 1/6 of the poor performances that the Jets have allowed, for example. A significantly strong QB will make his SOS look especially bad.

Moleculo -- this is inaccurate.http://www.pro-football-reference.com/blog/wordpress/?p=37

Please read that post, and tell me if you think a significantly strong team will make their SOS look especially bad (ignore the 6 weeks vs. 17 weeks distinction, because that's irrelevant). This uses the *exact* same formula.

no it's not - I stand by my statement. The 6 week vs 16 week distinction is not irrelevant - Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that. It's a sample size issue. 1/16 (or 1/8, for divisional opponents) isn't very good either. The fundamental flaw is that there simply aren't enough games to be able to apply something like correctly.It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

Any SOS-based normalization has to be taken with a grain of salt, and the fewer games played the less precise the model is likely to be.

Let me also add that this SOS discussion is very important because this is what allows Chase to draw the distinction between Brady and Garrard. Without normalizing for SOS, Brady's AYPA is >10% higher than Garrards, and therefore crowning Garrard as MVP is ludicrous. Chase is misapplying a mathematical tool to make an assumption which does not pass the common sense test.

Yup. And, while SOS may help put some added texture to raw statistics, it should by no means interpreted with the sort of confidence Chase applies to it. Especially after only 6 weeks, with one of those weeks (likely) affected by the statistics Brady and Garrard respectively contributed to their opponents SOS rankings.I appreciate Chase looking outside the box. I just think he's just misapplying some important assumptions here.

This is inaccurate.I don't understand how you'd look at six weeks worth of data and not do SOS adjustment. It's significantly more important to do after six weeks than after 16 weeks (do you understand why?). The way the SOS adjustment has been done here, Brady and Garrard's performance does not contribute to their opponent's SOS rankings.

Before I respond, would indulge me in just how you are computing SOS here and provide some actual raw data for the JAX and NE opponents. There are two issues at play here, but I am making some assumptions on how you're calculating the data. I just want to be sure I understand what you're doing with those data before offering a response.

Please see Doug's post. He did a terrific job explaining it.

Ok, I missed that earlier. I like the adjustment.Still, I think the inference is off. I'm thinking of a scenario here whereby you could have a QB be flawless and still rank lower than Garrard. And, when I say flawless, I mean flawless. This hypothetical QB could play all 16 games, complete 100% of his pass attempts, throw 6 TDs every game, throw for 7680 yards in a season and have zero interceptions and still rank lower than Garrard on your MVP list (and probably lower than David Carr, for that matter).

This hypothetical QB could line up at center and drop back to pass 120 times per game and complete 4-yarders on every play. First down: 4yd completion; second and 6 to go. Second down: 4yd completion; third and 2 to go. Third down: 4yd completion; 1st and 10. This QB has such uncanny accuracy that he never misses. He drives his team down the field on every possession and scores every time. It's that automatic. At the end of every game, he accounts for 6 TD passes and 480 yards passing on his 120 attempts.

If I understand your bias towards the AY/A, he would finish the season with a mark of 4.5 {[(7680 PaYds + (96TDs*10)) / 1920 attempts] = 8640/1920 = 4.5 AY/A}. This would be -1.47 below the league average (5.97). No matter what adjustments you made for SOS, he would still be ranked lower on your assessment than virtually all QBs in the game.

And, yet, in my mind, this hypothetical QB would be so off-the-charts in terms of his accuracy and the contributions he brings to his team that he would be the slam dunk of all slam dunks for MVP.

Brady isn't that QB. But, what he is doing is in a different realm than what Garrard is, who by his own right, is contributing so much to his team's success thus far. While I think statistical analyses can/should guide interpretation(s), they miss important variables along the way. We do what we can to reduce those misses, but in large part, I think your current MVP analysis misses some important qualitative features that Brady is contributing above/beyond Garrard.

Again, I do like your analysis, and it gives us a peek into their respective contributions (adjusting for the level of competition, to boot). But, at the end of the day, I don't think the conclusion you've reached is accurate.

Brady is the clear-cut MVP, thus far.

Brady's AY/A is higher than any QB has ever had since the merger. AY/A boosts Brady, it doesn't hurt him. His AY/A is slightly higher than Manning's was in 2004.

My point was that AY/A (or RAY/A) is a neat stat, but it is incomplete, no matter who your QB MVP is.

 

[Doug Drinen]

Still wondering why 10 yards per TD pass is right. I see a basis for the 45 yards per interception. Did I miss the explanation on the TD pass value?

You didn't miss it. It hasn't been given in this thread. The original research was done by Pete Palmer, Bob Carroll, and John Thorn in a book called The Hidden Game of Football. I don't want to do it the injustice of a quick summary, so instead I'll just try to convince you that it's in the right ballpark by presenting you with this hypothetical situation....It's the first game of the season, the first drive of the game. Your team is on offense. You and your opponent are evenly matched. You have first-and-ten at your own 30 yard line. Which would you rather see on the next play?

1. A 69-yard pass that puts you at 1st-and-goal at the one,

or

2. A 70-yard TD pass, followed by a 15-yard excessive celebration penalty (so you'll kick off from the 15 instead of the 30)?

If you have to think about it --- if it's not immediately clear --- then you must think that the marginal value of the TD, compared to a same-yardage pass that is not a TD is worth approximately 15 yards.

Obviously, which of those two options you'd choose depends on the score, the time remaining in the game, and how you match up with your opponent in certain facets of the game. But try this experiment: take every TD pass you see this weekend and ask yourself, "if I could trade that TD for the ball at the one and 10 yards of field position on my opponent's next possession, would I do it?" I don't know if 10 is exactly the right answer, but I don't think it's much bigger than that.

Obviously some TDs are worth more than others and some yards are worth more than others, so any static TD-to-yard conversion is going to be too high in some situations and too low in others. But if we want a quick summary stat that relies on total numbers, we need a number. In my mind at least, the above hypo makes ten seem pretty believable.

 

[Tau837]

Still wondering why 10 yards per TD pass is right. I see a basis for the 45 yards per interception. Did I miss the explanation on the TD pass value?

You didn't miss it. It hasn't been given in this thread. The original research was done by Pete Palmer, Bob Carroll, and John Thorn in a book called The Hidden Game of Football. I don't want to do it the injustice of a quick summary, so instead I'll just try to convince you that it's in the right ballpark by presenting you with this hypothetical situation....It's the first game of the season, the first drive of the game. Your team is on offense. You and your opponent are evenly matched. You have first-and-ten at your own 30 yard line. Which would you rather see on the next play?

1. A 69-yard pass that puts you at 1st-and-goal at the one,

or

2. A 70-yard TD pass, followed by a 15-yard excessive celebration penalty (so you'll kick off from the 15 instead of the 30)?

If you have to think about it --- if it's not immediately clear --- then you must think that the marginal value of the TD, compared to a same-yardage pass that is not a TD is worth approximately 15 yards.

Obviously, which of those two options you'd choose depends on the score, the time remaining in the game, and how you match up with your opponent in certain facets of the game. But try this experiment: take every TD pass you see this weekend and ask yourself, "if I could trade that TD for the ball at the one and 10 yards of field position on my opponent's next possession, would I do it?" I don't know if 10 is exactly the right answer, but I don't think it's much bigger than that.

Obviously some TDs are worth more than others and some yards are worth more than others, so any static TD-to-yard conversion is going to be too high in some situations and too low in others. But if we want a quick summary stat that relies on total numbers, we need a number. In my mind at least, the above hypo makes ten seem pretty believable.

Thanks. It's funny, I posted a quote from the same book earlier about the 45 yards per interception, but I guess it's been long enough since I read that book that I don't remember the yards per TD discussion.

 

[NE_REVIVAL]

I've given lots of statistical arguments for why Garrard has been great this year, but I'll write something here to appease the other crowd. FWIW, it's important to remember that I don't think Garrard is anywhere near as good as Brady or Manning, and he's not going to play anywhere near as well as Brady or Manning the rest of the way. What he has done, however, is play just as well as them so far this year. (If we were to start the season over again, I don't think Garrard would play this well again, and Jacksonville would probably be 3-2 or 2-3).

Week 1: The Jaguars allow 246 rushing yards to Tennessee, while Jaguars RBs rush for 48 yards. Up 10-6 in the second half, Jacksonville can't try a 36 yard field goal due to an injury to their kicker. Late in the game, with Jacksonville down three, Marcedes Lewis runs into MJD, and Drew fumbles inside the TEN 10 yard line. Garrard throws a 47 yard TD to John Broussard, and throws for 200 yards and no interceptions against a Titans pass defense that made Peyton Manning and Drew Brees have their worst games of the season. Tennessee so far this season, has been a top three pass defense.

Week 2: Against Atlanta, Garrard throws for 272 yards on 25 pass attempts. Garrard was 7-of-8 for 122 yards on the two fourth-quarter scoring drives. He also completes four passes of over 20 yards.

Week 3: Against Denver, a team that had been playing by far the best pass defense in the NFL so far, Garrard completes 70% of his passes in an efficient performance with 1 TD and 0 INT. He also runs for 52 yards. On 3rd and 4 from the Denver 20 late in the game, Garrard runs for 19 yards.

Week 5: Against Kansas City, Garrard averages over 8 yards per pass, chipping in a TD and 0 INT. The Jaguars beat Kansas City in Arrowhead, where they've won 14 of their last 16 games outside of this one. Garrard ices the game with a 40 yard pass in the 4th quarter, followed by his TD throw. This Chiefs pass D would intercept Carson Palmer twice, and 7.4 yards per pass, the following week.

Week 6: Garrard throws for over 200 yards and 2 scores, beating Houston. He also runs for 31 yards. Jacksonville finally beats Houston, a team they've struggled against in the past.

Meanwhile, look at who Garrard is doing this with: Reggie Williams, Dennis Northcutt, MJD and Marcedes Lewis are the only receivers with 100 yards on the year. Jacksonville's receivers are obviously way worse than Indianapolis', Dallas' or New England's wide receivers, but Jacksonville is still 4-1. The only game they lost was when their defense got shredded, their kicker got hurt, and they had a bad fumble late in the game. Garrard has had a QB Rating over 100 in four consecutive games.

Garrard isn't that good. But he's been playing out of his mind good.

As Paul Simon said "A man hears what he wants to hear and dis-regards the rest".There is no doubt that Gararrd has played well and is worthy of at least some MVP consideration, but to elevate him you are cherry picking stats and factors you like and ignoring important ones that don't fit your hypothesis. You would never know it by reading what you have written, but Jax's sucess is hardly all about Gararrd. They are #2 in scoring defense and #4 in rushing offense and the one game where they played poorly in those areas they lost while scoring only 10 pts at home. Granted TEN does have an excellent defense, but Gararrd was not a difference maker. Contrary to what you may believe JAX has a great offensive line and while his rec aren't great, they do include 2 1st rd draft picks and MJD who is a hell of a player.

There are oodles of stats where Gararrd doesn't come close to measuring up against the likes of Brady and Manning. For instance, Gararrd is getting sacked aprox every 8 times he drops back to pass; this is an incredibly poor ratio, especially when you consider Jax is #4 in rushing offense. Brady & Manning get sacked aprox every 34+ times they drop back to pass and while some of that can be attributed to having a better OL, much of it is due to the obvious fact that Brady & Manning are much better at reading defenses and getting rid of the ball. Maybe in time Gararrd will learn to be better at this or maybe he never will ala Drew Bledsoe, but imo it is one of the big things that seperates the good qb's from the great ones.

Put Manning or Brady on Jax and they are likley undefeated or certainly have at least as many wins. Put Gararrd on NE or INDY and they likely have at least one loss. Put another way; there are probably a handful of QBs who could have lead JAX to their 4-1 record but very few (imo) who could do likewise for INDY and NE.

FWIW, having said all that, I still think Garrard & the Jags have a real good shot to beat the Colts on Mon.

 

[MDSkinner]

As a Bears fan it almost makes me want to punch myself for saying this, but I think you can make a better argument for Favre than you can for Garrard.

Garrard has played incredibly well, and managed the game perfeectly but with a strong running game and defense to back him up.

Favre has single-handedly carried that offense. He like Garrard also has a good defense, and he has better receivers, but he has done more for the success of GB than Garrard has for the success of JAX. JMO.

 

[Statorama]

Tom Brady. 21 TDs and 2 ints. Let's not overcomplicate it folks. Brady has been nothing short of amazing. Hands down, walkaway winner if there's ever been one.

Nobody hates the Patriots more than I do, but I have to say that Brady is just on another level right now. He's like a football Terminator.

 

[Irish]

BTW--Happy birthday, Chase.

Get off his jock. TIA

 

[Chase Stuart]

Tom Brady. 21 TDs and 2 ints. Let's not overcomplicate it folks. Brady has been nothing short of amazing. Hands down, walkaway winner if there's ever been one.

Nobody hates the Patriots more than I do, but I have to say that Brady is just on another level right now. He's like a football Terminator.

For sure. The Pats are unstoppable and have a great chance of going 16-0, and Brady's got a great chance to break the TD record.But what Garrard's done has really flown under the radar, IMO.

 

[moleculo]

Brady so far accounts for 1/6 of each of his opponents games. There is no getting around that.

It really doesn't matter how you spin it - Brady has lit up everyone he has faced so far. The flip-side of that coin is that every team he has faced so far has had at least one game with 3+ TDs allowed. Therefore, every team's defensive YPA allowed is worse than it would be had you excluded what Brady has done, and the SOS Brady has faced is therefore less than what it would have been had they not played him.

There is a way around the chicken-egg problem. It's called iteration. It's pretty darn cool. Let me walk you through it and try to explain why it does indeed factor out the situation you're concerned about.STEP 1: compute an unadjusted rating for every QB and for every defense.

At this point, Brady's rating is too high (because it hasn't been adjusted to account for the weak defenses he's played). Meanwhile, assuming they've faced a more-or-less average collection of QBs aside from Brady, the Bills, Jets, Cowboys, etc have defensive ratings that are too low (because they haven't been adjusted to account for the strong QBs (i.e. Brady) that they've faced).

STEP 2: adjust every QB's rating based on the average defense he has faced.

At this point, Brady's rating is likely closer to being "correct," but is probably too low because we've used the STEP 1 defense ratings, which (probably) underrate the defenses Brady has faced. This is your main point: that we have overcorrected Brady's numbers. And if we stopped here, we would indeed have overcorrected (probably).

STEP 3: now adjust each defense's rating based on the average QB it has faced.

Again assuming the Patriots' opponents have faced a fairly average collection of QBs, except for Brady, this will improve the rating of each of those defenses by the full amount by which Brady is better than average.

At this point, it's pretty subtle trying to figure out whether Brady and/or his opponents have been over- or under-corrected, but the next steps will take care of it.

STEPS 4 through 1000: repeat steps 2 and 3 over and over again. Adjust Brady again based on the new defensive estimates, and adjust the defenses again based on the new QB estimates. Eventually, the numbers will stop changing.

Let me show you how it works specifically in this case:

Brady's nominal AYPA is +3.298 yards per attempt above average. That's what the raw stat line says.

Nominally, the defenses he has faced look like this:

Jets: -1.61 (<---- 1.61 yards per pass WORSE than average)

Chargers: -.57

Bills: -.00

Bengals: -.54

Browns: -.73

Cowboys: +.73

Now, if you take a weighted average of those six numbers, weighted by how many attempts Brady had against each team, you get -.36. So in the first pass, we'd decrease Brady's rating from +3.298 to about +2.94. [NOTE: if we stopped here, your criticism would be dead on. That's why we're not stopping here ] Meanwhile, the defenses that faced Brady get a collective boost. Here's where they stand after one step:

Jets: -1.48

Chargers: -.40

Bills: +.97

Bengals: +.28

Browns: -.55

Cowboys: +.31

Note how all the defenses that played against Brady are better than they appear (except the Cowboys, who have faced Bulger, Edwards, Green, and Grossman). This is your point moleculo, and it's a good point, which is why we're not finished yet.

So after we re-adjust Brady based on those revised defensive numbers, Brady moves back up from +2.94 to +3.18. Then we just do the whole thing over and over and over again. As the process moves on, Brady's ratings do the following:

3.30 ---> 2.94 ---> 3.18 ---> 3.04 ---> 3.14 ---> 3.07 ---> 3.12 ---> 3.09 ---> 3.12 ---> 3.11 ---> 3.12 ---> 3.13 ---> 3.13 ---> 3.13 ---> ...

And here are where the Pats' opponent's numbers eventually settled in:

Jets: -1.26

Chargers: -.27

Bills: +1.19

Bengals: -.21

Browns: -1.20

Cowboys: +.60

Note that each of those defenses has been adjusted by the full amount that the QBs they've faced are better or worse than average. For example, look at the Bills.

The Bills have faced:

Brady (+3.13)

Romo (+1.63)

Cutler (+.84)

Big Ben (+.54)

Pennington (+.12)

The weighted average of those is +1.20, which is exactly the amount that the iteration moved the Bills' rating. Remember, the stat line said the Bills are exactly average (-.00). After iteration, we find that Bills' schedule has been 1.2 better than average, so they end up with a rating of 1.2, which is the number we're ultimately using to adjust Brady.

Here is another way to see it:

The raw stats said Brady's schedule was .36 worse than average. After iteration, we figure that it's .16 worse than average. So the iteration bumped these teams up by a collective .20. Let's examine where that came from. For starters, Brady is 3.13 better than average. Now, if you go look at the *other* QBs faced by teams faced by Brady, you'll find that they are, on (weighted) average, about .38 worse than average (I can list them and their ratings if you want). Since about 1/6 of the schedule was against Brady, he bumped their ratings by about 3.13*(1/6). Meanwhile, the other QBs, which account for 5/6 of their collective schedule, bumped them down by about .38*(5/6). Since 3.13*(1/6) - .38*(5/6) is just about .2, that's why Brady's schedule got a boost of .2 for playing against Brady, along with a collection of other QBs that were a little below average.

It's all extremely self-referential (the chicken-egg thing your referred to earlier). It is, in a sense, infinitely self-referential. But it all balances. I think it's pretty cool.

EDIT: Just to pre-empt a potential hijack....

I used AYPA in the above example, but all of the above is agnostic about whether AYPA is worth anything or not. [For the record, I happen to like it, but I don't have as much energy as young Chase, so I'll let him answer those things]. This post is only about the SOS adjustment. You could do the same with (almost) any stat.

Doug,This procedure gets around the chicken-egg problem, but it doesn't address my two prime concerns - (1) any player who's performance is significantly different than the average will effect the SOS scores of his opponents, and (2) Too few games to make any SOS arguement useful.

(1)

I think that this algorithm does wonders for most of the population you are examining, but I think it breaks down at the extreme points. If someone performs consistently different than normal, that persons performance would have an effect on the SOS. Note: I say "different" instead of "better", because the same effect happens at the bottom of the rankings as well.

Here are some numbers I posted earlier: Jets have allowed 1460 yards, 10 TD's and have 4 pics, on 187 attempts. Outside of the NE game, they have allowed 1163 yards, 7 TD,s and 4 pics, on 165 attempts. The AY/A = 7.38 overall, but 6.38 vs QB's not named Brady. The Browns have given up 1581 yards, 17 TD's, 5 ints, on 233 attempts, for AY/A of 6.55. Vs Non Brady teams, they have given up 1316 yards, 14 TDs and 5 ints, in 211 attempts, for an AY/A of 5.83. So - Outside of the NE game, the Jets are a full point better and Cleveland is almost 3/4 of a point better. My point is that Brady's SOS correction value would be different had his performance been excluded from the SOS calculation in the first place.

In a nutshell, it's the self-referential bit I have a problem with.

(2)

the effect of a players impact on the SOS he faced will be higher with a relatively small number of games. Like I said earlier, Brady accounts for 1/6 of who his opponents have faced. The more games played, the magnitude of this effect gets reduced and SOS becomes more meaningful. In my opinion, SOS really cannot be in the discussion at all, until ~10 games are played. Again, I have no data to back that up.

I have a couple of solutions which could make the SOS algorithm better.

(a) When you are calculating SOS for Brady and you are looking at his opponents, take the Patriots game out of the equation completely. Break that circular reference. Don't let a team's performance bias the results of the SOS, which is what is happening now. Repeat for all teams. This becomes more tedious, but I think it's the right thing to do. If you are doing this ranking in excel, well it gets tough but if you are using some sort of programming language (Matlab, Maple, or other), it wouldn't be terrible to do.

(b) toss out the outliers when calculating a teams scores. Toss out the best and worse performance, toss out anything more extreme than a sigma from their normal, something. That would take care of this problem, as Brady's numbers would be tossed out. I think this might be a good thing to do regardless.

 

[Chase Stuart]

Brady is ridiculous. He is the real MVP, although Garrard's a fun underdog to back.

 

[Doug Drinen]

(1) any player who's performance is significantly different than the average will effect the SOS scores of his opponents

Let me take one more stab at convincing you that what you're saying here is not true. I ran the first-six-weeks numbers again under two alternative scenarios,

SCENARIO #1: I removed the week 5 New England / Cleveland game altogether. Like it never happened.

SCENARIO #2: I gave Brady 50 extra yards in the New England / Cleveland game. Instead of 265/3 on 38 attempts, I gave him 315/3 on 38 attempts.

Under scenario #1, the Browns D has a rating of -1.99.

Under scenario #2, the Browns D has a rating of -1.45.

In other words, giving up 315 yards and 3 TDs (and no INTs) against Brady helps the Browns' defensive rating (as well it should). It doesn't hurt it.

 

[Finless]

Brady is ridiculous. He is the real MVP, although Garrard's a fun underdog to back.

I agree that a case could be made for Brady although I'm not sold yet.

 

[Sheriff66]

This thread should be :lock:

 

[Doug Drinen]

Brady is ridiculous. He is the real MVP, although Garrard's a fun underdog to back.

I agree that a case could be made for Brady although I'm not sold yet.

Now that's funny.

 

[cobalt_27]

Brady is ridiculous. He is the real MVP, although Garrard's a fun underdog to back.

Huh? Just yesterday, you were certain that your data supported Garrard. What is it about today that changed anything? Brady's been doing this all year.

 

[Chase Stuart]

Brady is ridiculous. He is the real MVP, although Garrard's a fun underdog to back.

Huh? Just yesterday, you were certain that your data supported Garrard. What is it about today that changed anything? Brady's been doing this all year.

It's all in good fun. I think Garrard's done a great job, but I think Brady's pulled ahead after a 6 TD performance today.

 

[Pat Patriot]

Wow.

1. Lots of credit being given here for "thinking outside the box". Taking statistics and molding them in a way that convinces you that David Garrard is a more deserving MVP candidate at this point in the season than Tom Brady is just a waste of time and waste of space on this site.

2. If Chase has EVER been accused of being a Pats homer, that is a very uninformed reader/poster.

3. What is wrong with watching the games and letting that tell you who the best players are? I feel the same way about Hall of Fame voting. Its not about the statistics. You should be able to recognize greatness when you see it.

4. What is wrong with Wins, TD's, Yardage, Completion percentage. These are the statistics I have been using, along with my personal observations, for over 25 years and they have served me right.

5. I'm to be led to believe that Garrard has faced a tougher schedule than Brady because he has played Tennessee, Atlanta,

Denver, KC and Houston? Not a lot of world beaters there. At least the Pats have destroyed SD and Dallas.

Whoever posted it is right, what changed your mind about Brady from today's peformance. I'm sure Chase would have rated his performance today below average because it happened against a poor Miami D. Adjust for SOS and Brady is no better than Jay Cutler.