So, I was on my way to work when I thought about 2011-2012 and how almost every team seemed like a potential playoff contender. Unless you were the Browns, the "Any Given Sunday" philosophy could be posited on any team.
Flashforward to 2019 and it feels anything but. You have some teams tanking, like the Dolphins, but then teams like the Giants who are taking 3-4 years to turn around the ship.
I'm starting to believe parity is dead in the NFL due to certain teams having an advantageous upper hand in the analytics departments, especially with utilizing talent compared to their salary cap hit. Before, teams would go through the cycle of being good for a few years and then see a downturn. Now, teams have figured out how to stockpile talent cheaply while paying their perennials superstars (See Dallas Cowboys).
Am I going crazy or does this seem at least somewhat the case?
The more I think about the Giants, the more I'm starting to realize this current NFL does not allow for quick turnarounds like before. Ten years ago, any team with a few good drafts could rise to the top. Now, it feels if you've had few bad years of drafting followed by good ones, you're less likely to be able to get out of the dumpster fire of the NFL.
A specific example?? No. It's almost midnight and my brain isn't a fucking computer.
You've been watching the Andy Reid coach football for the last 2 decades though right? You're so dug in that you won't conceed that Andy Reid is garbage at clock management?
I'd also add, the discussions regarding 'analytics' supporting certain patterns of decision making by the Giants speaks nothing to the underlying decision making driving that decision, which is what I'm curious about. Shurmur might just naturally be aggressive in certain situations more than most coaches that lead to those results. Again, I'm not sure why looking at the resumes of guys allegedly leading the analytics efforts is frowned upon. Data science is a hard skill set to learn and master. Comparing the NYG 'analytics roster' to the Pats/Steelers/Eagles (for example) seems like an easy (but admittedly potentially flawed) way to see how deep the Giants are in verse their competitors.
I'm not a pom pom guy either, there are areas where the current coaching staff and FO are and have been failing IMO. I'm also under no delusion that they've been torch bearers in the recent wave of analytics...but there are some signs of life there if you bother to look.
It is pretty hard to convince me that DG adheres to analytics since he A) eschewed them publicly then B) drafted a RB with the 2nd pick of the draft.
But I was happy he finally selected a QB at #6 this year to get the rebuild underway. That's a plus. But he stuck with his aging QB too long, so there's that.
Sometimes they work in your favor but over the longitudinal view they do not. On field performance, we’ve been awful for a while now, probabilities not working in our favor overall for sure
C'mon man. Being awful on the field is due to a multitude of factors, such as drafting terribly and having a below average OL.
How in the hell does that relate to probabilities? If we make the correct decisions and fail - it isn't the decision itself that is always the root cause. What analytics people are saying is that by the math, the Giants are making sound decisions in-game. I'll repost this even though you'll ignore it:
Adherence to math and getting the right result are two different things. You either know that and willingly ignore it or you don't know that.
Neither is a good look.
What I've advocated for time and time again is that the Giants bring in someone that has experience engineering predictive systems, with education in math or deep experience in computer science and/or software engineering.
To you and Aces point i'm not arguing with the Giants attempting to integrate analytics i'm saying that if you are going to do it you should have top people ingrained in your organizations leadership with the proper experience. There isn't even any evidence that Gettleman is open minded enough to new ideas that if they hired someone it would work but back to why in a practical sense the above quote doesn't matter.
Again, the Giants don't show they have the game theory understanding to handle simple clock management but the 4th down or 2 pt conversions are more complex game theory than timeout usage. Haven't built these models but have thought through the proper architectures but a lot of 4th down math leans towards the fact that short yardage conversions are high % plays but this isn't in a vacuum. I'd rank conversion success and variables to raise the probability of that conversion in the following order of likelihood convert at higher rates.
1) Strong offensive line
2) Mobile QB
3) Talented RB
4) TE's with strong blocking AND receiving skills
5) Big physical WRs
Now reviewing this list, in terms of what the Giants were dealing with last year i'd say we were fairly weak in everything but the 3rd most important factor.
This is why employing and having more advanced models that like i'm saying again, the Pats clearly have someone making auto-encoding algorithms that turns video into physics equations that can be used in machine learning. That's an even better way to figure out the probability of conversion success the general force players play with and that force / change in force on recent plays. We can't even get basic software engineering to get clock management right.
In summation all this quote proves is we have people without the proper qualifications attempting to apply data that you need a TEAM of more qualified people to scratch the surface on how to apply properly in individual situations.
You know what is the funniest / stupidest part of this? You continually shit on PFF despite the fact that i've pointed out they have many people with the proper qualifications to pull off these kinds of calculations. You talk about how individual teams don't like the grades, guess what? I promise you the Giants aren't grading every player on every team on every play which is pretty much a fundamental step in starting an effective predictive system. Sure the teams have the calls but the fact that PFF doesn't have the play calls they still are trying to make a system that functions in light of that and it takes time to keep iterating and improving on that. The Giants are more than late in starting the process in earnest with a real technologist thinking about integrating software and advanced math into practical decision making.
You don't understand how this works. And it appears that the Giants don't either. And my biggest point is unlike say Kansas City where there might only be a few people that understand these systems and how to build them NYC is LITTERED with these people and we still don't have software to solve the SIMPLE problems like clock management.
Quote:
An analytical study pinpointed by Ben Baldwin on Twitter suggests the Giants were the most forward-thinking franchise in the NFL when it came to going for it on fourth down under head coach Pat Shurmur. The analytics dictate certain situations where the Giants and any other NFL offense should opt to go for it on fourth down. The Giants adhered to these analytics more than almost any NFL team in 2018.
What I've advocated for time and time again is that the Giants bring in someone that has experience engineering predictive systems, with education in math or deep experience in computer science and/or software engineering.
To you and Aces point i'm not arguing with the Giants attempting to integrate analytics i'm saying that if you are going to do it you should have top people ingrained in your organizations leadership with the proper experience. There isn't even any evidence that Gettleman is open minded enough to new ideas that if they hired someone it would work but back to why in a practical sense the above quote doesn't matter.
Again, the Giants don't show they have the game theory understanding to handle simple clock management but the 4th down or 2 pt conversions are more complex game theory than timeout usage. Haven't built these models but have thought through the proper architectures but a lot of 4th down math leans towards the fact that short yardage conversions are high % plays but this isn't in a vacuum. I'd rank conversion success and variables to raise the probability of that conversion in the following order of likelihood convert at higher rates.
1) Strong offensive line
2) Mobile QB
3) Talented RB
4) TE's with strong blocking AND receiving skills
5) Big physical WRs
Now reviewing this list, in terms of what the Giants were dealing with last year i'd say we were fairly weak in everything but the 3rd most important factor.
This is why employing and having more advanced models that like i'm saying again, the Pats clearly have someone making auto-encoding algorithms that turns video into physics equations that can be used in machine learning. That's an even better way to figure out the probability of conversion success the general force players play with and that force / change in force on recent plays. We can't even get basic software engineering to get clock management right.
In summation all this quote proves is we have people without the proper qualifications attempting to apply data that you need a TEAM of more qualified people to scratch the surface on how to apply properly in individual situations.
You know what is the funniest / stupidest part of this? You continually shit on PFF despite the fact that i've pointed out they have many people with the proper qualifications to pull off these kinds of calculations. You talk about how individual teams don't like the grades, guess what? I promise you the Giants aren't grading every player on every team on every play which is pretty much a fundamental step in starting an effective predictive system. Sure the teams have the calls but the fact that PFF doesn't have the play calls they still are trying to make a system that functions in light of that and it takes time to keep iterating and improving on that. The Giants are more than late in starting the process in earnest with a real technologist thinking about integrating software and advanced math into practical decision making.
You don't understand how this works. And it appears that the Giants don't either. And my biggest point is unlike say Kansas City where there might only be a few people that understand these systems and how to build them NYC is LITTERED with these people and we still don't have software to solve the SIMPLE problems like clock management.
you're describing every team in the NFL other than the Patriots and the Eagles. We are only a few years away from every team employing an analytics coach who either stands next to the HC or is up in the booth advising on time outs, 2-pt conversions, field position, win probability, etc etc. It is inevitable.
Fans talk shit necause they don't have to make the calls themselves.
Fans talk shit necause they don't have to make the calls themselves.
maybe so but most HCs and GMs are also ill-equipped for what is coming. Ex-jocks are generally not what you want running a team when the science, math and analytics takes over. The best HCs of the last 30 years, Walsh and Belichick are more scientists than jocks. It will soon be time to take the reigns away from ex-players who don't have the intellectual capacity for the change coming and recruit and train geniuses instead.
You know what really is the stupidest part of this? PFF's Player ratings go against almost everything analytics stands for.
- Subjective analysis
- Unqualified reviewers
- Questionable methodology
They aren't doing calculations!! They are subjectively grading a player each down and are doing such a poor job at it, they oftentimes aren't even directionally correct.
How can you be a supposed analytics "expert" and put weight into PFF? It is mind-boggling. It's like a scientist believing the Earth is flat.
Quote:
You know what is the funniest / stupidest part of this? You continually shit on PFF despite the fact that i've pointed out they have many people with the proper qualifications to pull off these kinds of calculations. You talk about how individual teams don't like the grades, guess what? I promise you the Giants aren't grading every player on every team on every play which is pretty much a fundamental step in starting an effective predictive system.
You know what really is the stupidest part of this? PFF's Player ratings go against almost everything analytics stands for.
- Subjective analysis
- Unqualified reviewers
- Questionable methodology
They aren't doing calculations!! They are subjectively grading a player each down and are doing such a poor job at it, they oftentimes aren't even directionally correct.
How can you be a supposed analytics "expert" and put weight into PFF? It is mind-boggling. It's like a scientist believing the Earth is flat.
from what i know about PFF they are analytically inclined but they aren't the beginning and end of analytics. They are trying to remake player evaluations mapping every play and pouring those figures into advanced models. That is very much an analytical approach. But they may not have the best approach or there may be a lot to critique about it. Just because they are imperfect doesn't mean they are not in the realm of analytics. It is not binary like that.
no they are not.
if you can do better i would encourage you to enter the business. There is a fortune to be made.
if you are correct, that they aren't qualified to grade film, that is an awfully easy thing to fix. It doesn't in any way speak to their business model. They just need to adjust their grading system or hire better graders. It has nothing to do with their business model. If you feel you know better I would urge you to enter the business. There is a lot of money to be made.
It's not a con to attempt to come up with better metrics and analytics for sports. The same exact things were said about Bill James and as it turns out he revolutionized the sport . He should be given his own wing in the Hall of Fame. And coaches, executives and fans alike vilified him as a con man throughout the 80s and 90s.
If PFF really wanted credibility instead of serving as stat porn for fans deluded into thinking they are getting fed statistics they would partner with the NFL, sit down with the teams and grade film with the necessary knowledge to do such a thing.
What they are doing now isn't just useless, it can't even serve a basis for future analysis. Pouring bad data into future predictive models nullifies any credibility.
he claims the Giants don't have the right people to do analytical calculations, but in the next breath, he claims that PFF does have the right people with the right credentials.
So what does he do? He ignores the information about the Giants utilizing analytics for in-game decisions, yet regales the board about PFF doing things the right way. and stands by the ratings process.
If PFF really wanted credibility instead of serving as stat porn for fans deluded into thinking they are getting fed statistics they would partner with the NFL, sit down with the teams and grade film with the necessary knowledge to do such a thing.
What they are doing now isn't just useless, it can't even serve a basis for future analysis. Pouring bad data into future predictive models nullifies any credibility.
yea.. you may be right. i have found some of their stuff confusing and bewildering but i admit i have not done a deep dive into what they are doing. But the idea of improving how players are analyzed/graded on a per play basis is excellent and whether PFF masters it or it falls to someone else, it is inevitable. Football needs to close the gap on baseball. It is easy to analyze every single play in a baseball game. Football is more confusing so there is no model yet to do it.
It's the appearance of credibility they're after.
he claims the Giants don't have the right people to do analytical calculations, but in the next breath, he claims that PFF does have the right people with the right credentials.
So what does he do? He ignores the information about the Giants utilizing analytics for in-game decisions, yet regales the board about PFF doing things the right way. and stands by the ratings process.
All we know for now is that analytics in football is in its infancy and the only people who seem to know what they're doing work for the Patriots and the Eagles. Beyond that I don't think we can say much with any certainty.
he claims the Giants don't have the right people to do analytical calculations, but in the next breath, he claims that PFF does have the right people with the right credentials.
So what does he do? He ignores the information about the Giants utilizing analytics for in-game decisions, yet regales the board about PFF doing things the right way. and stands by the ratings process.
Per my post you just don't really comprehend how these systems work or what their foundations are on.
You have where the Giants are.
1. No evidence of having anyone with advanced mathematics, computer science or software development skills.
2. Where some NFL teams are (some of these hires and Github libraries, the Giants don't have a github libarry)
3. Hybrid system like PFF where larger teams of people with the skills of #1 combine with a system like grading players
4. Purely quantitative system from converting film to physics equations or ZEBRA data
Now what you are failing to understand is a value based or outcome analysis systems could actually benefit from using data from kinds 3 and 4 because like i've said many times people and machines see different things and you can actually benefit from uniting the process and putting as much information of these different views in the hands of smart people.
Now you are trying to pretend like anything that touches subjective inputs is useless in machine learning? What dense person you are. Analyst estimates and price are valuable inputs in quantitative analysis and complex finance modeling EXACTLY because human input is layered in and it provides different signaling than pure data. Both are helpful.
2 sigma even got stopped from giving analysts surveys to basically layer in their personalities in addition to the data they provided to further signal off the estimates they gave.
You don't know anything about data analysis really yet you've also made yourself the arbiter of what is legitimate data and isn't? Give me a break, you know very little and continue to prove that
and walks were not considered important and they scoffed at OBP, for just about 20 years.
Quote:
PFF is a barometer to show just how dug in NoGainDayne is to his rant on analytics.
he claims the Giants don't have the right people to do analytical calculations, but in the next breath, he claims that PFF does have the right people with the right credentials.
So what does he do? He ignores the information about the Giants utilizing analytics for in-game decisions, yet regales the board about PFF doing things the right way. and stands by the ratings process.
Per my post you just don't really comprehend how these systems work or what their foundations are on.
You have where the Giants are.
1. No evidence of having anyone with advanced mathematics, computer science or software development skills.
2. Where some NFL teams are (some of these hires and Github libraries, the Giants don't have a github libarry)
3. Hybrid system like PFF where larger teams of people with the skills of #1 combine with a system like grading players
4. Purely quantitative system from converting film to physics equations or ZEBRA data
Now what you are failing to understand is a value based or outcome analysis systems could actually benefit from using data from kinds 3 and 4 because like i've said many times people and machines see different things and you can actually benefit from uniting the process and putting as much information of these different views in the hands of smart people.
Now you are trying to pretend like anything that touches subjective inputs is useless in machine learning? What dense person you are. Analyst estimates and price are valuable inputs in quantitative analysis and complex finance modeling EXACTLY because human input is layered in and it provides different signaling than pure data. Both are helpful.
2 sigma even got stopped from giving analysts surveys to basically layer in their personalities in addition to the data they provided to further signal off the estimates they gave.
You don't know anything about data analysis really yet you've also made yourself the arbiter of what is legitimate data and isn't? Give me a break, you know very little and continue to prove that
i think you can make your point without the derogatory tone. Subjective inputs on a play brings the fielding numbers like range factor to mind. I believe they were useful. And i think subjective determinations on a per play basis could be useful as well. Though they would need to be standardized. And yes integration with advanced next gen NFL metrics which are not subjective would seem to be a useful way forward.
It's not analytics, it's not real filmwork, it's not anything of value.
And it is telling that PFF is used by the majority of the teams (I believe 20+) for quantitative data.
How many use their ratings? Zero.
I know Sneakers understands this - others seemingly do not, but subjectivity isn't PFF's sin.
Combining subjectivity with unqualified reviewers, using a flawed methodology is their sin.
And it is a gigantic one. But the guy who supposedly knows analytics doesn't get that.
Do you understand they integrate subjective information?
Or are you disputing if they are useful in machine learning?
It isn't solely subjectivity in dispute. It is the combination with a flawed methodology and unqualified reviewers and operating with incomplete and incorrectly interpreted data.
Not only is the data better in doing price estimates, but the analysts are trained and considered experts in their field.
Now maybe you ignore points because you are too stupid to grasp them or you do it intentionally because you can't run roughshod over them in trying to praise PFF.
Neither one is a good look - but you do it over and over again.
Can you immediately become a price analyst by signing up on a website without any qualifications needed?
And I hope we were both wrong about Daniel Jones...
Can you immediately become a price analyst by signing up on a website without any qualifications needed?
this sounds so much like the model employed by Stats Inc 30 years ago. They hired barely qualified people, and by hired I mean didn't pay, to sit in the press box and do advanced charting of baseball games. My buddy would call me from the press box in the Big O as he charted games and then I would get calls from some bean counter there asking me who called me. But it's the same model. The point is, I think though, is that they are doing it, and if they modify and refine their practices they will become a big player in something that is inevitable.
Strategic moves are transparent when they happen. A player's responsibility is apparent. You don't have an endless array of formations. You don't have audibles.
A guy thows a ball, another one attempts to hit it and fielders attempt to catch it.
Data is much more straightforward and crisper. It is a matter of charting things, not making significant subjective decision.
Strategic moves are transparent when they happen. A player's responsibility is apparent. You don't have an endless array of formations. You don't have audibles.
A guy thows a ball, another one attempts to hit it and fielders attempt to catch it.
Data is much more straightforward and crisper. It is a matter of charting things, not making significant subjective decision.
Oh yes this is totally true. Baseball is essentially a game of binaries. But I would say two things. 1) Just because football isn't binary like baseball doesn't mean charting every play isn't useful and won't eventually yield a system or series of metrics that are better than what we have today, because what we have today is essentially garbage. 2) Baseball moved onto more complicated and subjective metrics such as range factor and other fielding metrics. Those were considered useful enough that the entire sabermetrics community accepted the conclusion that Derek Jeter was a lousy shortstop. That was based entirely on subjective inputs on a per play basis and it was accepted as truth. And I would bet the inputs were done by people no more qualified than the folks PFF employs. All this is to say that PFF may or may not be doing a good job, I just havent done the deep dive, but it could turn into something useful, even essential. I wouldn't dismiss their effort out of hand. Analytics needs to revolutionize football. Better data, better metrics, better thinking. And right now this is one of the only games in town.
Building a data driven org is not about following a punch card, it's not about having staff, it's not about "doing analytics."
The primary questions I ask when building a program 1) are you willing to fail 2) are you willing to hurt feelings.
If an org isn't willing to do those two things game over.
The second series is 1) are willing to let analysis guide all investment 2) are you willing to let data guide all operations. Again if the answer is no game over.
I've followed football keenly the last 10 or so years from a data perspective -- my feelings on the brand names are well known in the data field. Scoring player performance will take the next step in predictive motion analysis (play execution). That will be a step up from predictive scenario analysis (play calls) which isn't impossible now and is much more interesting than situational analysis (more on this later). And much easier is resource allocation analysis.
Doing some of this and calling it a day is a false sense of security and frankly more dangerous than doing nothing. If you follow a chart of what to do in scenario X, but don't know what to do against Y, you might as well go with your gut.
I'm thoroughly unimpressed by the Giants going for 2 points or passing more. That's kindergarten. They need to understand what plays are likely to succeed based on their roster. Following a punch card is "doing" analytics.
Same goes for investment. Like I posted above, I'd love to spar with the analyst who said Manning, Solder, Jenkings, and frankly Barkle fit a value-based investment system.
Employing thoughtful analysis then needs to be backed by results. If your models predict success, and you suck, self assess.
A ball travels a specific distance at a specific velocity. A batter pulls a curveball a specific % of the time. The right term is "charted", because a baseball play can be easily placed into a chart and categorized.
PFF's player ratings don't do that. They are trying to ascertain success or failure without knowing the assignment or the objective. They aren't charting plays - they are grading them - and that's where the significant difference comes into play
A ball travels a specific distance at a specific velocity. A batter pulls a curveball a specific % of the time. The right term is "charted", because a baseball play can be easily placed into a chart and categorized.
PFF's player ratings don't do that. They are trying to ascertain success or failure without knowing the assignment or the objective. They aren't charting plays - they are grading them - and that's where the significant difference comes into play
gotcha.. yes that does seem like a weird methodology.
George Box