The Agressive Decision Maker

[The Cooler King]

Former Bears coach and LA Charges head coach ruffling a lot of NFL discourse right now!

Where do you fall on the agressive/analytic verse traditional approaches to 4th down conversions and 2 point attempts?


I love it. Give me a Staley or Harbaugh or Chico type who isn't afraid to make a tough call and play to win. Give me the coach who puts it in his offenses hands within the 5 yard line. The guy who does it consistently and doesn't let one failure dictate the next decision. And none of these guys are super flashy new age guys in other ways. None are hot O coordinators. They strike a beautiful balance of incorporating analytic decisions into their culture and why it's about Ws for them.

That Staley is also the same guy who can sit there and talk about tangible benefits of why the run game is important to him and what factor it actually plays for him and his offense... It's honestly horrible, with hindsight the Bears let him get away. Per Kruetz on a recent podcast, Staley did not initially want to follow Fangio to Denver. He only left after Nagy went and got Pagano. Maybe we could have dumped Nagy and promoted Staley after 2020 if we kept him here as DC.

Anyways, would love you hear yalls thoughts on the decision making by Staley last night.

[Grizzled]

I'm not an analytics guy but the broadcasters were saying on at least several of the five 4th down plays, it was in the Chargers favor to go for it. We salute Staley (I think he's doing a great job as coach) but his decisions cost the Chargers the win. I have a problem in that the Chargers didn't even try to put the ball in the end zone at the end of the 1st half; all plays were short of the the goal line. It comes down to gut feelings and confidence in your team's abilities. I also like a coach going for it.

[The Marshall Plan]

I would imagine that in order for analytics to be effective for that decision the data would need a shit load of attributes. Like you can't just say something like the Bears have a 57% chance of converting on 4th down therefore go for it.

Not only are you looking at the probability of just the situation in isolation (4th and 2 let's say) but also:

Away or home
What yard line are you on?
Weather
Game momentum
Opposing defense ranking against the type of play you're going to call (run or pass)
Who the opposing QB is? Are you trying to keep The GOAT off the field?
The current score
How much time is left in the game?
If this play costs you the game, are you out of the playoff hunt?

There's a serious rabbit hole one could go down here analytically.

[The Cooler King]

I would imagine that in order for analytics to be effective for that decision the data would need a shit load of attributes. Like you can't just say something like the Bears have a 57% chance of converting on 4th down therefore go for it.

Not only are you looking at the probability of just the situation in isolation (4th and 2 let's say) but also:

Away or home
What yard line are you on?
Weather
Game momentum
Opposing defense ranking against the type of play you're going to call (run or pass)
Who the opposing QB is? Are you trying to keep The GOAT off the field?
The current score
How much time is left in the game?
If this play costs you the game, are you out of the playoff hunt?

There's a serious rabbit hole one could go down here analytically.

I mean the math behind the decision is pretty simple after you've chosen a win probability model. The models themselvds, can vary greatly as far as external inputs (other than down/distance/etc).

Pro football references for example basically just uses a Vegas spread. More complex models use actual opponent data, offense stats, defense stats, etc.

Analytical folks I follow on Twitter are pointing out Staley hasn't necessarily been only "by the models" this season. It's very likely LA has their own unique model and it's very likely the fact they were playing the Chiefs, probably the best team in league at erasing deficits, played a role in him deciding to try and maximize scoring in that situation, including going for it on their side of the field. Last night showed there wasn't a ton of difference between handing Mahomes the ball deep verse handing it over on a short field.

Favoring analytics should just be seen as favoring info and not being bound by one thought process. The math on these decisions are often very narrow and football is a game that's susceptible to very volatile swings. Its a gut punch to be agressive and feel it cost them, but there's no singular key play. Lots of individual decisions that are basically coin flip odds could have pointed to a different outcome.

[dplank]

I like the approach and generally like the idea of adding analytics, but will also say you don’t need an analytics guy to tell you to go for it on 4th and inches down 11 in the 4th. That was as bad as Fox challenging his own TD and having it taken off the board.

Because Bears

[The Marshall Plan]

I would imagine that in order for analytics to be effective for that decision the data would need a shit load of attributes. Like you can't just say something like the Bears have a 57% chance of converting on 4th down therefore go for it.

Not only are you looking at the probability of just the situation in isolation (4th and 2 let's say) but also:

Away or home
What yard line are you on?
Weather
Game momentum
Opposing defense ranking against the type of play you're going to call (run or pass)
Who the opposing QB is? Are you trying to keep The GOAT off the field?
The current score
How much time is left in the game?
If this play costs you the game, are you out of the playoff hunt?

There's a serious rabbit hole one could go down here analytically.

I mean the math behind the decision is pretty simple after you've chosen a win probability model. The models themselvds, can vary greatly as far as external inputs (other than down/distance/etc).

Pro football references for example basically just uses a Vegas spread. More complex models use actual opponent data, offense stats, defense stats, etc.

Analytical folks I follow on Twitter are pointing out Staley hasn't necessarily been only "by the models" this season. It's very likely LA has their own unique model and it's very likely the fact they were playing the Chiefs, probably the best team in league at erasing deficits, played a role in him deciding to try and maximize scoring in that situation, including going for it on their side of the field. Last night showed there wasn't a ton of difference between handing Mahomes the ball deep verse handing it over on a short field.

Favoring analytics should just be seen as favoring info and not being bound by one thought process. The math on these decisions are often very narrow and football is a game that's susceptible to very volatile swings. Its a gut punch to be agressive and feel it cost them, but there's no singular key play. Lots of individual decisions that are basically coin flip odds could have pointed to a different outcome.

Like @dplank said just now, some situations are common sense. You don't need a computer for it.

I'd be really interested in knowing how that data is setup and what those models look like. In order for that to be right those formulas have to be awfully complex.

Then what is the real world application?

You have 10 seconds to call the play. The app on the tablet they use would have to instantaneously know all these variables and make a decision. You don't have time to input all the variables. The computer could do it if it's setup right and there's some sort of API like connection with real time game data and the app.

The really interesting thing would be for an AI to start calling plays for everything. Not just 4th down.

[dplank]

I know the Ravens employ a big Harvard data/analytics guy and were ahead of the curve on that (as they usually are). Another aspect of many why big money in ownership matters beyond the overly simplistic "salary cap" constrained POV. They were early on it, and didn't mind paying some nerd 200K or whatever just to see if it added any value or not. The Bears are dinosaurs who look to copy other people's success instead of creating their own.

Because Bears.

[Moriarty]

Aggressive because it's ballsy/manly/exciting/whatever (like Pace's big trades) sucks.

Aggressive because the analytics say so, sure.

[dplank]

Then what is the real world application?

You have 10 seconds to call the play. The app on the tablet they use would have to instantaneously know all these variables and make a decision. You don't have time to input all the variables. The computer could do it if it's setup right and there's some sort of API like connection with real time game data and the app.

The really interesting thing would be for an AI to start calling plays for everything. Not just 4th down.

Really smart thought, I went down this road once with a guy who works at USA Football and wanted to create a dynamic playsheet driven by algorithms (I don't like to call things AI that aren't actually self learning, but it's AI-ish). The idea was that you'd have the call sheet loaded onto a tablet, and position coaches have tablets also. The position coaches have one job, to feed their best guess at how each player in their position group is performing relative to their primary assignment (i.e., is the LT beating the DE, what degree of confidence do you have that he will beat the DE on a run block, on a pass block, rated from 1-10). All that data across every position goes against the playbook and call sheet, updated in real time as position coaches are constantly feeding their ratings (might change mid game if there's an injury, for example) and plays get highlighted to the play caller based on down/distance/score/situation and then layering in probabilities that take in every players potential win% and their role in the play.

It was a fascinating project that sadly never got off the ground.

If anyone has a spare 500K we could probably launch a business around this. I'm tapped trying to get another business off the ground right now lol - too many ideas and not enough time lol.

[Otis Day]

I didn't have a problem with any of Staley's 4th down gambles. If your team can't make 3rd and 4th and shorts, you probably don't deserved to win the game anyway. Against KC, it is hard to beat them with FGs.

[The Cooler King]

I know the Ravens employ a big Harvard data/analytics guy and were ahead of the curve on that (as they usually are). Another aspect of many why big money in ownership matters beyond the overly simplistic "salary cap" constrained POV. They were early on it, and didn't mind paying some nerd 200K or whatever just to see if it added any value or not. The Bears are dinosaurs who look to copy other people's success instead of creating their own.

Because Bears.

Haha exactly. I saw one Bears fan trying to dunk on Staley last night and I had to call them out.

[The Cooler King]

I mean the math behind the decision is pretty simple after you've chosen a win probability model. The models themselvds, can vary greatly as far as external inputs (other than down/distance/etc).

Pro football references for example basically just uses a Vegas spread. More complex models use actual opponent data, offense stats, defense stats, etc.

Analytical folks I follow on Twitter are pointing out Staley hasn't necessarily been only "by the models" this season. It's very likely LA has their own unique model and it's very likely the fact they were playing the Chiefs, probably the best team in league at erasing deficits, played a role in him deciding to try and maximize scoring in that situation, including going for it on their side of the field. Last night showed there wasn't a ton of difference between handing Mahomes the ball deep verse handing it over on a short field.

Favoring analytics should just be seen as favoring info and not being bound by one thought process. The math on these decisions are often very narrow and football is a game that's susceptible to very volatile swings. Its a gut punch to be agressive and feel it cost them, but there's no singular key play. Lots of individual decisions that are basically coin flip odds could have pointed to a different outcome.

Like @dplank said just now, some situations are common sense. You don't need a computer for it.

I'd be really interested in knowing how that data is setup and what those models look like. In order for that to be right those formulas have to be awfully complex.

Then what is the real world application?

You have 10 seconds to call the play. The app on the tablet they use would have to instantaneously know all these variables and make a decision. You don't have time to input all the variables. The computer could do it if it's setup right and there's some sort of API like connection with real time game data and the app.

The really interesting thing would be for an AI to start calling plays for everything. Not just 4th down.

I think I saw something about Detroit having one guy who's primary responsibility is feeding that data to the coaches. Doesn't mean they have to use it, but they're constantly aware. Key to this process is this person is giving them projections even on 2nd or 3rd down for when it might be 4 down territory because then it opens up your playbook on those downs. Real smart to do it that way. You don't have to box yourself into a spreadsheet, but you can increase flexibility.

[The Cooler King]

I know the Ravens employ a big Harvard data/analytics guy and were ahead of the curve on that (as they usually are). Another aspect of many why big money in ownership matters beyond the overly simplistic "salary cap" constrained POV. They were early on it, and didn't mind paying some nerd 200K or whatever just to see if it added any value or not. The Bears are dinosaurs who look to copy other people's success instead of creating their own.

Because Bears.

I've seen some indication they've made improvements here at least. They're just coming from so far behind though. Hopefully this failure keeps pushing them. Failure seems to have pushed the Browns to innovate and invest.

[Mikefive]

I'm more conservative when it comes to these gambles. But there's times to do it and times to take the almost certain 3 points.

One important point that hasn't been brought up is matchups, which would be incredibly difficult to measure with analytics. If you have superior matchups in important areas (Davante Adams vs. Eddie Jackson in the slot, for example), then maybe you should go for it more often. But if your matchups don't favor you (See Bears offense vs. most of the league), then taking the risk is more of a losing proposition.

[IE]

Analytics (stats) are great if they're used properly and understood. One problem I have with last night is I think he fell into some Gambler's Fallacy & assumed that since they didn't make the previous one somehow they would be more likely to make it the next time. That's incorrect. Of course I don't know that he fell into that - but I highly suspect it. Because in addition to the probability of success they obviously HAVE to consider the opportunity cost of not having that 3 points.

To me there is a big difference between "the odds" and "the total risk outside of that given play". Situationally, in football there has always been a generally recognized thinking of "take the points early; go for it later if your situation is such that you need it". Early in the game, you can't really be sure how many possessions you'll get or how the game will unfold. But if you can score 3 points a possession with an NFL average of over 10 possessions per game, you're going to win more often than not - you take the points early. Later in the game you have a LOT better visibility into how many points you're looking to need to win - which better justifies the risk of not taking that 3 points - because you need more to win. Outside of what the analytics tell you about the likelihood of success on a given play, some external variables like the list TMP posted need to be considered.

I didn't like that aggression last night. It was stubborn, Tin Cup-type aggression. You want to be aggressive? How about just turning up your aggression in play calling overall a bit, and not swing for all or nothing points on a small number of plays? I believe Staley got caught up in being a modern analytics-driven coach but lost track of context and as a consequence lost the game.

[dplank]

I'm more conservative when it comes to these gambles. But there's times to do it and times to take the almost certain 3 points.

One important point that hasn't been brought up is matchups, which would be incredibly difficult to measure with analytics. If you have superior matchups in important areas (Davante Adams vs. Eddie Jackson in the slot, for example), then maybe you should go for it more often. But if your matchups don't favor you (See Bears offense vs. most of the league), then taking the risk is more of a losing proposition.

did you, ummmm, read the post above? Matchups are key to your decision making, i wrote a whole nerdy thing about it

[The Cooler King]

Aggressive because it's ballsy/manly/exciting/whatever (like Pace's big trades) sucks.

Aggressive because the analytics say so, sure.

Agreed totally. Blind agressiveness is just a gambler mentality. It should have backing.

I think any good coach will try to use the model and numbers and weigh the human element of the team their leading. If your players think your just gambling away chances they are gonna lose faith and the results may be impacted. But I'm sure a lot of it is expectation setting and framing. Get buy in as a leader!

[HisRoyalSweetness]

I didn't have a problem with any of Staley's 4th down gambles. If your team can't make 3rd and 4th and shorts, you probably don't deserved to win the game anyway. Against KC, it is hard to beat them with FGs.

The one that stood out that I thought was a bad decision was passing on the FG just prior to halftime which would have put the Chargers up by 7 with the Chiefs getting the ball first in the second half. In that situation it made more sense to me to play the percentages and extend your lead to a TD.

[The Marshall Plan]

Just to keep this discussion going because the idea of analytics or A.I. in football is really interesting to me.

Supercomputers nowadays spank world class master level chess players.

Why couldn't Skynet For Football do the same thing to opposing defenses?

Now one thing to consider is in chess the move is executed perfectly. Like it's not hard to move a knight piece to it's appropriate position.

But in the NFL there are multiple human based points of contact that can fail. Although I suppose the A.I. could be programmed to account for that in it's modeling.

Why not in 20 years, or even much less time than that, couldn't Skynet take over playcalling and relay the play to the QB?

[dplank]

I think the concern here is that Skynet ultimately realizes that the fans are not a 'value add' to their core mission to win a game, and deploys microbots into the stadium to eradicate us all.

[HisRoyalSweetness]

Ooh, can the Packers trial this Skynet idea first?

[The Cooler King]

Just to keep this discussion going because the idea of analytics or A.I. in football is really interesting to me.

Supercomputers nowadays spank world class master level chess players.

Why couldn't Skynet For Football do the same thing to opposing defenses?

Now one thing to consider is in chess the move is executed perfectly. Like it's not hard to move a knight piece to it's appropriate position.

But in the NFL there are multiple human based points of contact that can fail. Although I suppose the A.I. could be programmed to account for that in it's modeling.

Why not in 20 years, or even much less time than that, couldn't Skynet take over playcalling and relay the play to the QB?

Interesting thoughts, but I guess besides the human execution issue is both sides are moving in simultaneous fashion. Chess is taking turns.

Now I'm sure early adopters could have their AI learn and master their human opponent pretty well in real time, but eventually it's just two AI bots trying to guess against each other once the late adopters join.

A very forward thinking commissioner might want to study and make rules preventing an potential total AI takeover to keep integrity of human element. But perhaps more art than science of what is an acceptable AI boundary or not.

[RichH55]

Skynet fixes games to about the same level as the Refs

[The Marshall Plan]

I think the concern here is that Skynet ultimately realizes that the fans are not a 'value add' to their core mission to win a game, and deploys microbots into the stadium to eradicate us all.

In that case I vote for the inaugural debut of Skynet to take place in Lambeau Field.

[IE]

Just to keep this discussion going because the idea of analytics or A.I. in football is really interesting to me.

Supercomputers nowadays spank world class master level chess players.

Why couldn't Skynet For Football do the same thing to opposing defenses?

Now one thing to consider is in chess the move is executed perfectly. Like it's not hard to move a knight piece to it's appropriate position.

But in the NFL there are multiple human based points of contact that can fail. Although I suppose the A.I. could be programmed to account for that in it's modeling.

Why not in 20 years, or even much less time than that, couldn't Skynet take over playcalling and relay the play to the QB?

Football "takes turns" too. The trick is the defense gets a chance to change the board after every move - and like you mentioned the range of potential outcomes versus the digital nature of a chess move. But there's no question someone could build a decision support system that could recommend the best move for each play, with options based on what the defense shows (and likelihood that is being disguised).

I like the idea of JF1 being the QB of the future, with a display overlayed onto his visor with real-time feedback (enemy proximity sensors, success probability metrics overlayed onto receivers in his view). Like if it is Allen Robinson, even if he looks open the display says "Abort" in red.... and it if is Jesper Horsted it says "Always Open" in green.

[The Marshall Plan]

I like the idea of JF1 being the QB of the future, with a display overlayed onto his visor with real-time feedback (enemy proximity sensors, success probability metrics overlayed onto receivers in his view). Like if it is Allen Robinson, even if he looks open the display says "Abort" in red.... and it if is Jesper Horsted it says "Always Open" in green.

YES!

Like some cross between The Terminator and Robocop.

[The Marshall Plan]

Just to keep this discussion going because the idea of analytics or A.I. in football is really interesting to me.

Supercomputers nowadays spank world class master level chess players.

Why couldn't Skynet For Football do the same thing to opposing defenses?

Now one thing to consider is in chess the move is executed perfectly. Like it's not hard to move a knight piece to it's appropriate position.

But in the NFL there are multiple human based points of contact that can fail. Although I suppose the A.I. could be programmed to account for that in it's modeling.

Why not in 20 years, or even much less time than that, couldn't Skynet take over playcalling and relay the play to the QB?

Football "takes turns" too. The trick is the defense gets a chance to change the board after every move - and like you mentioned the range of potential outcomes versus the digital nature of a chess move. But there's no question someone could build a decision support system that could recommend the best move for each play, with options based on what the defense shows (and likelihood that is being disguised).

I like the idea of JF1 being the QB of the future, with a display overlayed onto his visor with real-time feedback (enemy proximity sensors, success probability metrics overlayed onto receivers in his view). Like if it is Allen Robinson, even if he looks open the display says "Abort" in red.... and it if is Jesper Horsted it says "Always Open" in green.

This is going to radically change the game.

Can't you see somebody like a Jerry Jones literally spend like $10M (or whatever) on Skynet, where the salary cap does not apply nor do current league rules I believe, have Skynet tell the OC, then the OC tells the QB?

Then you get into the whole RIse Of The Machines thing. Competing software / hardware brands. Teams going batshit crazy upgrading their A.I. server with like 50 NVIDIA cards chain linked and shit.

It actually would be kinda cool if I didn't think it would totally take over the game and turn it into Madden 2100 with real people.

Skynet could then get into the whole Wargames thing (The Matthew Broderick movie.) where it just simulates an entire game over and over and over again. Concludes football is a useless exercise that can only wind up in a tie and stops the apocalypse.

[southdakbearfan]

Football "takes turns" too. The trick is the defense gets a chance to change the board after every move - and like you mentioned the range of potential outcomes versus the digital nature of a chess move. But there's no question someone could build a decision support system that could recommend the best move for each play, with options based on what the defense shows (and likelihood that is being disguised).

I like the idea of JF1 being the QB of the future, with a display overlayed onto his visor with real-time feedback (enemy proximity sensors, success probability metrics overlayed onto receivers in his view). Like if it is Allen Robinson, even if he looks open the display says "Abort" in red.... and it if is Jesper Horsted it says "Always Open" in green.

This is going to radically change the game.

Can't you see somebody like a Jerry Jones literally spend like $10M (or whatever) on Skynet, where the salary cap does not apply nor do current league rules I believe, have Skynet tell the OC, then the OC tells the QB?

Then you get into the whole RIse Of The Machines thing. Competing software / hardware brands. Teams going batshit crazy upgrading their A.I. server with like 50 NVIDIA cards chain linked and shit.

It actually would be kinda cool if I didn't think it would totally take over the game and turn it into Madden 2100 with real people.

Skynet could then get into the whole Wargames thing (The Matthew Broderick movie.) where it just simulates an entire game over and over and over again. Concludes football is a useless exercise that can only wind up in a tie and stops the apocalypse.

And then Bellicheck hacks said software in a skynetgate scandal in 2032.

[RichH55]

I know the Ravens employ a big Harvard data/analytics guy and were ahead of the curve on that (as they usually are). Another aspect of many why big money in ownership matters beyond the overly simplistic "salary cap" constrained POV. They were early on it, and didn't mind paying some nerd 200K or whatever just to see if it added any value or not. The Bears are dinosaurs who look to copy other people's success instead of creating their own.

Because Bears.

So being Smart matters? Ok. That I agree with (that's money ball actually - not money matters)

Which Owner couldn't afford 200K for the nerd?

Shad Khan - didn't go cheap to hire Urban Meyer. 10 Million a year for Urban Meyer. Bad. 200 K for Nerd. Good

But Big Money in Ownership matters?

Is that the point I'm reading?

[dplank]

Hmmmm. Odd timing/post weeks after this convo ended and moved into a running Terminator joke. Seems targeted.

Big money owners lay out 200K for a nerd not even knowing if he'll help or not, just on the hope that he may add value and if not, so be it. They lead vs follow. Small money owners set tight budgets for the team operators and enforce them aggressively. If a small budget team wanted to do something like this, they'd have to ask for budget in advance, get it approved, and roll it out - they'd have to prove why they need it. Meanwhile, the big budget team has a couple years ahead of integrating this stuff into their operation. They're always ahead of things, sometimes it works sometimes it doesn't - if money doesn't matter, then they're fine with it.

Having made a living selling into these teams personally, I can tell you there was a dramatic difference in how teams like the Cowboys operated vs teams like the Bears. The Bears operated like a small budget organization during my 7 years of doing business in the league. Baseball was same, the Yankees behaved fundamentally differently than the Orioles. Both could afford my software, but one took the leap immediately when they saw any possibility of competitive advantage, while the other waited for justification of the increased spend in their budget. Hockey - same. Hoops - same. Rugby/Soccer - same. I created my sales target list based on this perspective, which teams operated like the Cowboys vs which teams operated like the Bears, and applied my energy accordingly.

Eventually in MLB I sold almost every team. When I'd go to the winter meetings, the difference between the Yankees/Red Sox and Orioles/Brewers was astounding. The Yankees would roll in to town with a crew of like 30 people, Red Sox too. The Orioles, Brewers? 4 or 5. Could the Orioles owner have afforded to send 30? Absolutely yes. Did he? No. The Yankees would present me with a giant list of things they wanted us to build for them, and pay us to develop them. The Orioles would sit down and ask me, what's new coming out this year?

Just an analogy, but when you have worked directly with these teams as I have you see how differently the teams with insane money operate vs the teams that aren't at that level. Hell, Jerry Jones ultimately bought my entire company from me just so that he could steer the tech where he wanted to take it. They operate at a completely different level than the Bears.