Analytics

[Bear-Hawk]

They had an interesting discussion about “analytics” on GMF this morning. Computers are replacing coaches on 4th down decisions. The point Peter made (and I agree) is that the coach should consider how the game is going. Is my defense good enough to stop them? Is my offense playing well enough to score if we get the ball back? Because of analytics, teams are just going for it on 4th down far more now than in years past, where they should be punting to gain field position or kicking the damn field goal to tie the game.

It happened over and over yesterday, which is why they were talking about it on GMF.

 

[sutz]

The problem with analytics in football is the small sample sizes. Across 17 games (16 historically) at about 150 plays per game, there isn't enough data to really build statistical probabilities, so I agree with you. The coach should rely on his gut feel on the situation and the risk factors involved. Punting on 4th and 1 is probably the right choice from deep in your own end of the field in the 1st quarter, but if there is 2 minutes left in the 4th and you're trailing, probably not.

Baseball, with 162 games a year and literally thousands of plays and situations lends itself to analytics with much better probabilities of outcomes being predictable.

 

[Bear-Hawk]

The problem with analytics in football is the small sample sizes. Across 17 games (16 historically) at about 150 plays per game, there isn't enough data to really build statistical probabilities, so I agree with you. The coach should rely on his gut feel on the situation and the risk factors involved. Punting on 4th and 1 is probably the right choice from deep in your own end of the field in the 1st quarter, but if there is 2 minutes left in the 4th and you're trailing, probably not.

Baseball, with 162 games a year and literally thousands of plays and situations lends itself to analytics with much better probabilities of outcomes being predictable.

A bigger sample would reduce the sampling error of your estimate of what teams have done previously in the situation, but the coach is still getting paid to use his own intelligence and knowledge of how this particular game is going to make a decision.

 

[Bear-Hawk]

Holy cow. We had it again last night. Raiders went for 2 pt. Conversion and lost the game by 1 pt. They had over 4 minutes on the clock plus all 3 timeouts. That’s plenty of time to win the game. Make the other team have to score to win. KC only needed to make first downs to run out the clock and make Raiders spend their timeouts. Yes, I know the argument on the other side, but this was the wrong decision.
Screw the “analytics”!

 

[AgentDib]

Good comments, it really is about specificity. The actual probabilities vary based on the personnel on the field and how the game has unfolded thus far, and thus any generic "here's what the book says" advice sounds a bit silly. When talking about a particular play, the actual probability of conversion could be anywhere from 1%-99% depending on the players and play call and of course that completely changes the math.

Who is the best suited to estimate the likelihood of conversion? Probably the head coach, but it's fun to talk about. Broadcasts could be more interesting by solving the equation backwards to say something like "going for two makes sense here provided they are at least 75% likely to succeed."

 

[Maelstrom787]

Holy cow. We had it again last night. Raiders went for 2 pt. Conversion and lost the game by 1 pt. They had over 4 minutes on the clock plus all 3 timeouts. That’s plenty of time to win the game. Make the other team have to score to win. KC only needed to make first downs to run out the clock and make Raiders spend their timeouts. Yes, I know the argument on the other side, but this was the wrong decision.
Screw the “analytics”!

I don't think the analytics were necessarily behind those decisions, and they absolutely weren't for the punt that Reid decided on.

The models had the 2pt decisions as coin flips at best. The coaches did the dumb things, especially McDaniels

 

[Bear-Hawk]

I don't think the analytics were necessarily behind those decisions, and they absolutely weren't for the punt that Reid decided on.

The models had the 2pt decisions as coin flips at best. The coaches did the dumb things, especially McDaniels

The original Dick Vermeil chart says go for two points if trailing by 1 pt.

 

[RiverDog]

The problem with analytics in football is the small sample sizes. Across 17 games (16 historically) at about 150 plays per game, there isn't enough data to really build statistical probabilities, so I agree with you. The coach should rely on his gut feel on the situation and the risk factors involved. Punting on 4th and 1 is probably the right choice from deep in your own end of the field in the 1st quarter, but if there is 2 minutes left in the 4th and you're trailing, probably not.

Baseball, with 162 games a year and literally thousands of plays and situations lends itself to analytics with much better probabilities of outcomes being predictable.

That's just one problem with analytics. The other is that they are not team specific. If you're not a good short yardage running team or if your starting center is out, then your odds of converting on 4th and 1 may not be as good as what the book is telling you.

Analytics are a great tool for coaches to use as a reference when they need to make a decision. But they shouldn't be relied upon as gospel. The HC needs to take into account all possible information, including his subjective intuition, like the flow of the game, the look in his players eyes, weather conditions, or a black cat running across the field, when he makes his call.

 

[WmHBonney]

Analytics give a HC an excuse when he goes along with them and they don't work out. "Well, analytics said to go for it so we did."

 

[bmorepunk]

The problem with analytics in football is the small sample sizes.

The uncertainty could be pretty large, but the same could be said of the coaches. When you have the data for all recent NFL games compiled and you see how things go generically over time it's a lot more information than you have as a head coach, regardless of how many years you've been doing it.

Coaches who use data to gauge generic probabilities but use their own personal assessments (particularly those with a lot of experience and emotional control) probably do better than the other coaches. But if you just make choices based on data and expecting it to be deterministic? You're in trouble.

 

[HawkRiderFan]

I feel like analytics understates the negative outcome on decisions. Also, imo, you can't just use numbers to make your decisions when the outcome can be unpredictable due to the game played by humans who will make mistakes.

On that Ravens decision to go for it on 4th and goal, "they couldn't account for the int" that gave he ball to the Bills at the 20 and not the 3. Well that sort of thing needs to be factored in when making decisions.

The numbers can guide you but I think as some people have already said you have to factor in what's happening in the specific situation

 

[toffee]

The problem with analytics in football is the small sample sizes. Across 17 games (16 historically) at about 150 plays per game, there isn't enough data to really build statistical probabilities, so I agree with you. The coach should rely on his gut feel on the situation and the risk factors involved. Punting on 4th and 1 is probably the right choice from deep in your own end of the field in the 1st quarter, but if there is 2 minutes left in the 4th and you're trailing, probably not.

Baseball, with 162 games a year and literally thousands of plays and situations lends itself to analytics with much better probabilities of outcomes being predictable.

But not all guts were created equal?