4th down Bot - Good look at analytics

[T-Long]

Not sure if Marrone uses something like this...but it's really cool to look at. This is an analysis of every 4th down that the Bills had over the year, and if Marrone made the correct call going for it or kicking.

http://www.nytimes.com/newsgraphics/2013/11/28/fourth-downs/team.html?teamid=BUF

[Typ0]

Re: 4th down Bot - Good look at analytics

Interesting stuff. I would like to see the actual model they used. I toyed around with this stuff when I was more into the NFL ten years ago. Were I to have a steady up-datable stream of raw data I am confident my models would have been pretty good. Two huge downfalls of this study:

1) I do not see any input of environmental data -- things like wind, rain, snow, temperature, field condition, field type, etc. These are things that have considerable impact on the playing out of a game IMO and cannot be ignored. One of the coolest things I ever saw was that last TD we scored at the snow game VS. Miami. We went to kick - off and whamo the sky opened up and the persistent snow became an all out blizzard. Visibility was not conductive to throwing a pass LOL. On the other hand if you can connect on a timing route I doubt the defenders could even chase after the guy. Things like that not only effect the decision in question but make an impact on the data you are using for the analysis. These are highly correlated variables that render models insignificant if they aren't included.

2) The premise of the study is wrong. Coaches are not trying to maximize points. They are trying to maximize the difference between their score and the opponents score in that particular game. Thus, I think the Bot is an idiot. There really is no reason to punt at all and they put some 'dampening' factor in there to make teams punt so the public didn't dismiss the guy making these models as an idiot.

[Novacane]

Re: 4th down Bot - Good look at analytics

Coach's punt to often IMO but if there are as many stupid "I'd of gone for it" by bot as stupid decisions by Marone to punt.

Couple of examples 4th and 3 at our own 17, up by 7 against ATL with 4:38 to go in the game. Bot would of gone for it.

4th and 8 at our own 9 with 3:50 to go in the game down by 1 to Miami. Bot would of gone for it.

[Later]

Re: 4th down Bot - Good look at analytics

LOL at the idea that the least forward-thinking organization in the NFL would ever use something like this to guide decisions.

[ICRockets]

Re: 4th down Bot - Good look at analytics

Is this complete? I see no reference to the 4th and Goal vs Cincy.

[YardRat]

Re: 4th down Bot - Good look at analytics

LOL at 'too close to call' and 'it's complicated'.

[Mr. Pink]

Re: 4th down Bot - Good look at analytics

Not surprising a bot would say to go for every 4th and short on the field. Odds are something like 70% a team will convert a 4th and 1 and around 80% when they decide to run over 90% on a QB sneak.

I'm guessing the bot is programmed with percentages on what plays give you a better chance to win the game at the end of the day.

An example brought up is the 4th and 3 vs Atlanta...converting that makes the percentage of winning go up, punting makes the percentage go down...obviously going for it and not getting it also makes the percentage of winning go down but there is likely a better percentage gain of victory than a greater chance of losing if you don't make the first vs punting.

Actually in that circumstance since they're up 7, a TD only takes the game to overtime so I'll guess the percentage of losing that game via punting or getting stopped on 4th down is small.

[Mace]

Re: 4th down Bot - Good look at analytics

I waffle on analytics. On one hand I get the concept, on the other hand I can see how stats get skewed by capable people. Like if you have a juggernaut back, 4th and 1 is going to be a good bet, if you have a Peyton Manning, a certain percentage throw will be near a sure thing......so people who have that back or a Peyton are going to adjust the numbers by trying things others won't with a higher rate of success. You can look at percentages for kicks past a certain yardage that you can't try if you have a Ryan Lindell for less attempts, and make more of them with a Janikowski who will make more attempts.

I'd have trouble mathing a Marrone who had variable personnel week to week to mess up his maybes. But it's a fun link.

[better days]

Re: 4th down Bot - Good look at analytics

Well, you just have to look at the number of 4th downs the Bills faced to know why they went 6-10.

[Typ0]

Re: 4th down Bot - Good look at analytics

Not surprising a bot would say to go for every 4th and short on the field. Odds are something like 70% a team will convert a 4th and 1 and around 80% when they decide to run over 90% on a QB sneak.

I'm guessing the bot is programmed with percentages on what plays give you a better chance to win the game at the end of the day.

An example brought up is the 4th and 3 vs Atlanta...converting that makes the percentage of winning go up, punting makes the percentage go down...obviously going for it and not getting it also makes the percentage of winning go down but there is likely a better percentage gain of victory than a greater chance of losing if you don't make the first vs punting.

Actually in that circumstance since they're up 7, a TD only takes the game to overtime so I'll guess the percentage of losing that game via punting or getting stopped on 4th down is small.

No, the bot is considering the mixture of plays and decisions that lead to scoring the most points. It doesn't look at anything about winning or how many points the opponents score. It's right in the article.

[IlluminatusUIUC]

Re: 4th down Bot - Good look at analytics

2) The premise of the study is wrong. Coaches are not trying to maximize points. They are trying to maximize the difference between their score and the opponents score in that particular game. Thus, I think the Bot is an idiot. There really is no reason to punt at all and they put some 'dampening' factor in there to make teams punt so the public didn't dismiss the guy making these models as an idiot.

Of course there is. There is always the probability of conversion weighed against the probabilities of the opponents' scoring when then they get the ball back.


Personally, I agree that coaches punt far far too much. I would need a seriously long conversion to consider punting in opponent's territory and anyone who punts inside the opponent's 40 should be fired.

[Typ0]

Re: 4th down Bot - Good look at analytics

Of course there is. There is always the probability of conversion weighed against the probabilities of the opponents' scoring when then they get the ball back.


Personally, I agree that coaches punt far far too much. I would need a seriously long conversion to consider punting in opponent's territory and anyone who punts inside the opponent's 40 should be fired.

Again, the article clearly states the objective is to score the most points not to score more points than the opponent. There is a factor built in because when you punt you don't have the ball and your likelihood of scoring goes down significantly. But winning the game is not part of this model maximizing points scored it.