Took a look at 34 NFL QBs (cut-off was at least 1,000 passing yards season-to-date) across 7 different stats:
(1) QB Rating
(2) Completion Percentage
(3) Passing Yards per Attempt
(4) Passing Yards per Completion
(5) Passing Yardage per Game
(6) Interceptions per 100 Passes
(7) TD Passes to Interception Ratio
For each of the above statistic, I simply ranked all 34 QBs from 1 to 34. I then took the MEDIAN across all 7 statistical rankings, and then sorted all QBs on their 7-stat MEDIAN ranking from best to worst.
Importantly, this analysis does not include any rushing statistics, which Daniel Jones excels at. That being said, here is our QB's ranking for each statistic and then his final Median Rank:
Daniel Jones Ranking (out of 34 QBs)
QB Rating (17th)
Completion Percentage (14th)
Passing Yards per Attempt (20th)
Passing Yards per Completion (27th)
Passing Yardage per Game (28th)
Interceptions per 100 Passes (3rd)
TD Passes to Interception Ratio (10th)
Daniel Jones Final 7-stat Median Ranking: (17th)
And below is the Final 7-stat Median Ranking for all 34 QBs:
Jalen Hurts 3
Tua Tagovailoa 3
Patrick Mahomes 4
Geno Smith 6
Jimmy Garoppolo 6
Joe Burrow 7
Ryan Tannehill 8
Josh Allen 9
Justin Herbert 10
Jared Goff 12
Tom Brady 13
Andy Dalton 14
Trevor Lawrence 14
Derek Carr 15
Russell Wilson 16
Daniel Jones 17
Jacoby Brissett 17
Kyler Murray 18
Dak Prescott 19
Lamar Jackson 19
Kirk Cousins 20
Aaron Rodgers 21
Mac Jones 22
Matthew Stafford 22
Taylor Heinicke 22
Marcus Mariota 25
Carson Wentz 26
Davis Mills 27
Justin Fields 27
Matt Ryan 28
Cooper Rush 29
Baker Mayfield 30
Zach Wilson 31
Kenny Pickett 32
My bottom-line Daniel Jones take-away from these stats: Given the absolute decimation of his wide receiver unit, a final Median Ranking of 17 ain't too shabby. Add to that his rushing ability and I say the Giants have an above average QB in Daniel Jones.
In this case, you want a particular stat to spike (or drag) the composite score because it's very much part of that QB's performance identity. If a QB ranks highly across the board but has one particular stat that drags his whole average down, that also represents some sort of massive flaw in his game reflected in the statistical output.
You would want that flaw and corresponding drag on the QB's score to be factored in if you're trying to quantify a composite ranking of QBs. You've already eliminated noise by setting a fair threshold for qualification in the first place. Going the next step by using median rather than mean is now eliminating useful insight because the noise is already removed.
Hope that helps.
Imagine banging the table like Carl does, only for a middle third QB.
Quote:
In terms of average vs median, there is no "right" way, but I prefer the median to prevent any one particular stat ranking to pull a QB's final score either too low or too high. In any event, I went back to the data and used the average -- Daniel Jones final average was 17, which placed him #15 out of 34 QBs. That's practically the same vs the median score for him -- 17 (which places him #16 out of 34 QBs.)
In this case, you want a particular stat to spike (or drag) the composite score because it's very much part of that QB's performance identity. If a QB ranks highly across the board but has one particular stat that drags his whole average down, that also represents some sort of massive flaw in his game reflected in the statistical output.
You would want that flaw and corresponding drag on the QB's score to be factored in if you're trying to quantify a composite ranking of QBs. You've already eliminated noise by setting a fair threshold for qualification in the first place. Going the next step by using median rather than mean is now eliminating useful insight because the noise is already removed.
Hope that helps.
I’m not sure I quite follow you from a statistical/interpretive point of view, but in any event I think the median-average issue is sorta moot since both stats generate more or less the same ranking, especially in the case of Daniel Jones.
You can win with Daniel Jones under these 2 circumstances:
1. He's on a rookie deal
2. He has a great defense and excellent supporting cast around him.
If either of thosse 2 aren't true, you aren't winning anything.
Quote:
In comment 15935435 M.S. said:
Quote:
In terms of average vs median, there is no "right" way, but I prefer the median to prevent any one particular stat ranking to pull a QB's final score either too low or too high. In any event, I went back to the data and used the average -- Daniel Jones final average was 17, which placed him #15 out of 34 QBs. That's practically the same vs the median score for him -- 17 (which places him #16 out of 34 QBs.)
In this case, you want a particular stat to spike (or drag) the composite score because it's very much part of that QB's performance identity. If a QB ranks highly across the board but has one particular stat that drags his whole average down, that also represents some sort of massive flaw in his game reflected in the statistical output.
You would want that flaw and corresponding drag on the QB's score to be factored in if you're trying to quantify a composite ranking of QBs. You've already eliminated noise by setting a fair threshold for qualification in the first place. Going the next step by using median rather than mean is now eliminating useful insight because the noise is already removed.
Hope that helps.
I’m not sure I quite follow you from a statistical/interpretive point of view, but in any event I think the median-average issue is sorta moot since both stats generate more or less the same ranking, especially in the case of Daniel Jones.
Great thread M.S.! Thanks for posting.
Just wondering, after having gone through this exercise, in your opinion, if there were a factor(s) which you didn't include in your analysis, which would cause the final rankings to be significantly different what might it(they) be?
Quote:
RUSHING? That’s why he is around 12th best in league. You need to make it a separate category.
Imagine banging the table like Carl does, only for a middle third QB.
Dunk agree he is an average qb. That is with this team. If he had a Wr or two which didn’t have stone hands or could run after the catch he could probably fall to around #10. That’s where the numbers for Phil or Eli were. They were not rated higher vs their peers. He has the qualities and toughness that they had and we could win with him without question. A couple of skill receivers and LBs (is dire need) and we should be able to play with anyone. Different ways to build championship teams. Imagine if he had Chase and Tee Higgins? The list goes on and on. Even with the Jets receivers he could be ranked around #10.
2011 Newton, Locker, Gabbert, Ponder
2012 Luck, Griffin, Tannahill, Weeden
2013 Manuel
2014 Bortles, Manziel, Bridgewater
2015 Winston, Marriota
2016 Goff, Wentz, Lynch
2017 Trubisky, Mahomes, Watson
2018 mayfield, Darnold, Allen, Rosen, Jackson
2019 Murray, Jones, Haskins
2020 Burrow, Tua, Herbert, Love
Excuse the spelling
That’s a great question!
I would like to see a stat such as % completion rate
(1) under pressure
(2) when WRs are covered
(3) for a first down on a third down play
Also, yardage thrown based on only the ball in the air (excludes yards after catch.)
That’s a great question!
I would like to see a stat such as % completion rate
(1) under pressure
(2) when WRs are covered
(3) for a first down on a third down play
Also, yardage thrown based on only the ball in the air (excludes yards after catch.)
Forgot to say… I’m not sure how these would impact rankings.
Quote:
In comment 15935435 M.S. said:
Quote:
In terms of average vs median, there is no "right" way, but I prefer the median to prevent any one particular stat ranking to pull a QB's final score either too low or too high. In any event, I went back to the data and used the average -- Daniel Jones final average was 17, which placed him #15 out of 34 QBs. That's practically the same vs the median score for him -- 17 (which places him #16 out of 34 QBs.)
In this case, you want a particular stat to spike (or drag) the composite score because it's very much part of that QB's performance identity. If a QB ranks highly across the board but has one particular stat that drags his whole average down, that also represents some sort of massive flaw in his game reflected in the statistical output.
You would want that flaw and corresponding drag on the QB's score to be factored in if you're trying to quantify a composite ranking of QBs. You've already eliminated noise by setting a fair threshold for qualification in the first place. Going the next step by using median rather than mean is now eliminating useful insight because the noise is already removed.
Hope that helps.
I’m not sure I quite follow you from a statistical/interpretive point of view, but in any event I think the median-average issue is sorta moot since both stats generate more or less the same ranking, especially in the case of Daniel Jones.
The simplest way of explaining my point is that the very thing you're trying to remove (a singular statistical data point that disrupts a player's overall ranking by use of a composite average) is something that you should actively be including, because that disruptive data point is more likely to be a valid part of that QB's statistical identity than it is to be an outlier. Even if the results of the list are similar, the process would be more valid by use of mean than by median for what you're trying to demonstrate.
Part of the reason for this is that once you have narrowed the list to only the QBs who have reached a certain threshold of activity in order to qualify, you've already eliminated the chances of a misleading outcome in the data (e.g., a low-usage backup QB who happens to be leading the league in a few rate statistics based on a deceivingly low number of attempts).
Going a step further, because you're using rankings as your data itself (and not some sort of aggregate composition of the underlying football stats), you already have guardrails on the values. If there are only 34 qualifying QBs, every QB will have a value of 1-34 in each of the 7 categories. To isolate the median among those values results in using only one ranking (whichever lands 4th of the 7 chosen categories) of one statistic per QB. It's just not enough data, even if the results bear out similarly to using mean.
If I ask you what 2+2 equals, and you instead calculate 2^2, you will arrive at the same result as the correct answer, but you will have gotten there incorrectly and your accidentally correct answer would not be replicable with other values.
Passer rating is a flawed statistic largely because it is disproportionately informed by completion percentage. Using both here, along with another stat that is closely tied to completion percentage (Y/A), basically doubles down on the inherent flaw in that statistic, particularly since both are already included in the passer rating calculation to begin with.
You seem to have a keen interest in statistical storytelling, so I'm hoping you do find this feedback helpful.
Why Passer Rating Is Broken - ( New Window )
Quote:
That’s a great question!
I would like to see a stat such as % completion rate
(1) under pressure
(2) when WRs are covered
(3) for a first down on a third down play
Also, yardage thrown based on only the ball in the air (excludes yards after catch.)
Forgot to say… I’m not sure how these would impact rankings.
Thanks M.S. When I look at your list the thing that jumps out is Tua and Hurts. Two guys whose careers were in question last year. Now they're in the conversation for mvp. What changed? Both got a stud receiver to throw to.
Seems to point to the quality of a QBs supporting cast as a major if not the primary factor in a QBs performance.
Quote:
In comment 15935506 Gatorade Dunk said:
Quote:
In comment 15935435 M.S. said:
Quote:
In terms of average vs median, there is no "right" way, but I prefer the median to prevent any one particular stat ranking to pull a QB's final score either too low or too high. In any event, I went back to the data and used the average -- Daniel Jones final average was 17, which placed him #15 out of 34 QBs. That's practically the same vs the median score for him -- 17 (which places him #16 out of 34 QBs.)
In this case, you want a particular stat to spike (or drag) the composite score because it's very much part of that QB's performance identity. If a QB ranks highly across the board but has one particular stat that drags his whole average down, that also represents some sort of massive flaw in his game reflected in the statistical output.
You would want that flaw and corresponding drag on the QB's score to be factored in if you're trying to quantify a composite ranking of QBs. You've already eliminated noise by setting a fair threshold for qualification in the first place. Going the next step by using median rather than mean is now eliminating useful insight because the noise is already removed.
Hope that helps.
I’m not sure I quite follow you from a statistical/interpretive point of view, but in any event I think the median-average issue is sorta moot since both stats generate more or less the same ranking, especially in the case of Daniel Jones.
The simplest way of explaining my point is that the very thing you're trying to remove (a singular statistical data point that disrupts a player's overall ranking by use of a composite average) is something that you should actively be including, because that disruptive data point is more likely to be a valid part of that QB's statistical identity than it is to be an outlier. Even if the results of the list are similar, the process would be more valid by use of mean than by median for what you're trying to demonstrate.
Part of the reason for this is that once you have narrowed the list to only the QBs who have reached a certain threshold of activity in order to qualify, you've already eliminated the chances of a misleading outcome in the data (e.g., a low-usage backup QB who happens to be leading the league in a few rate statistics based on a deceivingly low number of attempts).
Going a step further, because you're using rankings as your data itself (and not some sort of aggregate composition of the underlying football stats), you already have guardrails on the values. If there are only 34 qualifying QBs, every QB will have a value of 1-34 in each of the 7 categories. To isolate the median among those values results in using only one ranking (whichever lands 4th of the 7 chosen categories) of one statistic per QB. It's just not enough data, even if the results bear out similarly to using mean.
If I ask you what 2+2 equals, and you instead calculate 2^2, you will arrive at the same result as the correct answer, but you will have gotten there incorrectly and your accidentally correct answer would not be replicable with other values.
I appreciate all your thoughts here and I think I have a better understanding of what you are driving at. I've reproduced your comments below and have added a few observations of my own:
"The simplest way of explaining my point is that the very thing you're trying to remove (a singular statistical data point that disrupts a player's overall ranking by use of a composite average) is something that you should actively be including, because that disruptive data point is more likely to be a valid part of that QB's statistical identity than it is to be an outlier."
What you say here is true, but you probably know as well as anyone that -- technically speaking -- a median statistic does not remove a particular observation in the calculation. It just tamps down the influence of an outlier observation and does so more effectively than what an "average" statistic can do. That said, I understand your thinking as to why this was not necessary in the first place and why I should have just used the average.
"Part of the reason for this is that once you have narrowed the list to only the QBs who have reached a certain threshold of activity in order to qualify, you've already eliminated the chances of a misleading outcome in the data (e.g., a low-usage backup QB who happens to be leading the league in a few rate statistics based on a deceivingly low number of attempts)."
Again, your statement here has a lot of merit. My original intent was to confine this analysis to starting QBs. Truth be told, I didn't even think about how the non-starting QB stats would have exhibited even greater variability, but I am certain you are correct about that. I guess where I'm a little confused is why that then necessitates that I stay away from a median stat for the starters. I understand there would be less variability with them, but I'm not sure it is thus cut-and-dried about "median" vs "average."
"Going a step further, because you're using rankings as your data itself (and not some sort of aggregate composition of the underlying football stats), you already have guardrails on the values. If there are only 34 qualifying QBs, every QB will have a value of 1-34 in each of the 7 categories. To isolate the median among those values results in using only one ranking (whichever lands 4th of the 7 chosen categories) of one statistic per QB. It's just not enough data, even if the results bear out similarly to using mean."
I definitely see where you are going here and I find myself agreeing with most of what you are saying. But is it true that the median I employed is only using one ranking? The median is choosing one ranking that falls exactly in the middle of the other six rankings... so is it really just one? More to the point, a middle/median ranking for one QB does not have to come from the same statistic as another QB. For example, Andy Dalton has a final median ranking of 14, which came from his yards/completion rank. Derek Carr has a final median ranking of 15 which comes from his yards/attempt rank.
"If I ask you what 2+2 equals, and you instead calculate 2^2, you will arrive at the same result as the correct answer, but you will have gotten there incorrectly and your accidentally correct answer would not be replicable with other values."
Again, you make a good point here. For what it's worth, I went back to the data and re-ranked all 34 QBs based on an average stat. I then correlated that with my original "median" stat. The r-squared value = 95.7%. Practically the same rank order with little variation between the two sets of data.
Passer rating is a flawed statistic largely because it is disproportionately informed by completion percentage. Using both here, along with another stat that is closely tied to completion percentage (Y/A), basically doubles down on the inherent flaw in that statistic, particularly since both are already included in the passer rating calculation to begin with.
You seem to have a keen interest in statistical storytelling, so I'm hoping you do find this feedback helpful. Why Passer Rating Is Broken - ( New Window )
You are absolutely correct that there is redundancy in some of the stats.
Quote:
In comment 15935628 M.S. said:
Quote:
That’s a great question!
I would like to see a stat such as % completion rate
(1) under pressure
(2) when WRs are covered
(3) for a first down on a third down play
Also, yardage thrown based on only the ball in the air (excludes yards after catch.)
Forgot to say… I’m not sure how these would impact rankings.
Thanks M.S. When I look at your list the thing that jumps out is Tua and Hurts. Two guys whose careers were in question last year. Now they're in the conversation for mvp. What changed? Both got a stud receiver to throw to.
Seems to point to the quality of a QBs supporting cast as a major if not the primary factor in a QBs performance.
I think you are dead-on about that. High end WRs can only make a QB better!
Quote:
No TDs and using median over mean. He would look even worse with these 2 modifications.
Couple of things:
(1) I'm assuming you mean TD passes. That's fair. Daniel Jones ranks 23rd in TD passes alone. But if I used that stat, I probably would have also used INTs, in which case Daniel Jones ranks 5th... so it's sort of a wash.
(2) In terms of average vs median, there is no "right" way, but I prefer the median to prevent any one particular stat ranking to pull a QB's final score either too low or too high. In any event, I went back to the data and used the average -- Daniel Jones final average was 17, which placed him #15 out of 34 QBs. That's practically the same vs the median score for him -- 17 (which places him #16 out of 34 QBs.)
Low INTs to low TDs is no a "wash." Teams need to score points to win football games, QBs need to throw the ball down the field to make it easier for the run game. DJ is 35th in air yards per attempt...which means this is a dinknand dunk offense designed to limit his turnovers. They don't trust him to play like a stud QB.
Quote:
In comment 15935260 Matt M. said:
Quote:
No TDs and using median over mean. He would look even worse with these 2 modifications.
Couple of things:
(1) I'm assuming you mean TD passes. That's fair. Daniel Jones ranks 23rd in TD passes alone. But if I used that stat, I probably would have also used INTs, in which case Daniel Jones ranks 5th... so it's sort of a wash.
(2) In terms of average vs median, there is no "right" way, but I prefer the median to prevent any one particular stat ranking to pull a QB's final score either too low or too high. In any event, I went back to the data and used the average -- Daniel Jones final average was 17, which placed him #15 out of 34 QBs. That's practically the same vs the median score for him -- 17 (which places him #16 out of 34 QBs.)
Low INTs to low TDs is no a "wash." Teams need to score points to win football games, QBs need to throw the ball down the field to make it easier for the run game. DJ is 35th in air yards per attempt...which means this is a dinknand dunk offense designed to limit his turnovers. They don't trust him to play like a stud QB.
You could be right they don’t trust Daniel Jones, but isn’t it just as plausible they don’t trust their non-existent WR unit and their o-line’s shaky pass blocking?