2009-10 Bears Season: Utler, 'Uther 'Uckers

[SKO]

I liked the challenge, the timeout there doesn't matter a ton and watching the play live I thought Forte was down... The only negative of that particular challenge is that the Bears would only have had one more challenge if the call were confirmed.

Also, if it was overturned.  Unless you're preemptively upset over the loss at a chance of getting a third challenge?  You're really quite ridiculous.

But he graduated from college, man.

[Dr. Nguyen Van Falk]

Just want to add the extra buzzkill that the point of Romer's paper was not necessarily to come up with The Ultimate Dialectically Perfect Socialist People's Historical Materialist Football Strategy.

Rather, the point was to analyze conservative, sub-optimal decision-making on the gridiron as a way of illustrating complications to the "central assumption of most economic models" that "firms maximize expected profits."

This paper shows that the behavior of National Football League teams on fourth downs departs from the behavior that would maximize their chances of winning in a way that is highly systematic, clear-cut, and statistically significant. This is true even though the decisions are comparatively simple, the possibilities for learning and imitation are unusually large, the compensation for the coaches who make the decisions is extremely high, and the market for their services is intensively competitive. Despite these forces, the standard assumption that agents maximize simple objective functions fails to lead to reasonably accurate descriptions of behavior.

That's some straight up faggot shit right there.

[Pre]

The problem with these kinds of studies is that they ignore the fact that 4th down
plays don't exist in a vacuum.  First off, much of NFL defensive strategy is predicated
on the fact that if you hold a team on 3rd down they are going to kick, this
frequently inflates the yardage on 3rd down when defenses give up the 4 yard check
down on 3rd and 8 or whatever.  In this genius' math, he uses 3rd down results to
predict 4th down results.  So the fact that defenses will usually let a team pick up
50 to 75% of 3rd and long influences his math to think that of course you go for it
on 4th down because you can get that 50 to 75% again.

The second thing these studies don't factor in is macro-game situations.  Like if you're
up by 2 late in the game, maybe if you go for it and get it then you greatly increase
your chance to get 7 more points and winning by 9, where if you fail then you greatly
increase you chance of losing by 1.  This is similar to all the anti-bunt numbers in
baseball.  All of them use run probability to conclude that giving up the out lowers
your chance at a big inning.  But in a lot of scenarios late in the game you don't need
a 7 run inning, you just need that 1 run, and you'll trade away the chance for a big
inning for a chance for 1.

Sure, some situations might call for more aggression on 4th down, but there's way too
many interconnected aspects of this to devolve it to a simple 'solve for x' kind of problem.

edit - Also, they don't factor in individual team's abilities.  If you have a bad ass defense and
a horrific o-line, you're more likely to benefit from kicking/playing conservative on offense vs
if you have a fantastic o-line and bad defense.

[SKO]

The problem with these kinds of studies is that they ignore the fact that 4th down
plays don't exist in a vacuum.  First off, much of NFL defensive strategy is predicated
on the fact that if you hold a team on 3rd down they are going to kick, this
frequently inflates the yardage on 3rd down when defenses give up the 4 yard check
down on 3rd and 8 or whatever.  In this genius' math, he uses 3rd down results to
predict 4th down results.  So the fact that defenses will usually let a team pick up
50 to 75% of 3rd and long influences his math to think that of course you go for it
on 4th down because you can get that 50 to 75% again.

The second thing these studies don't factor in is macro-game situations.  Like if you're
up by 2 late in the game, maybe if you go for it and get it then you greatly increase
your chance to get 7 more points and winning by 9, where if you fail then you greatly
increase you chance of losing by 1.  This is similar to all the anti-bunt numbers in
baseball.  All of them use run probability to conclude that giving up the out lowers
your chance at a big inning.  But in a lot of scenarios late in the game you don't need
a 7 run inning, you just need that 1 run, and you'll trade away the chance for a big
inning for a chance for 1.

Sure, some situations might call for more aggression on 4th down, but there's way too
many interconnected aspects of this to devolve it to a simple 'solve for x' kind of problem.

Hmm. This.

[Dr. Nguyen Van Falk]

The problem with these kinds of studies is that they ignore the fact that 4th down
plays don't exist in a vacuum.  First off, much of NFL defensive strategy is predicated
on the fact that if you hold a team on 3rd down they are going to kick, this
frequently inflates the yardage on 3rd down when defenses give up the 4 yard check
down on 3rd and 8 or whatever.  In this genius' math, he uses 3rd down results to
predict 4th down results.  So the fact that defenses will usually let a team pick up
50 to 75% of 3rd and long influences his math to think that of course you go for it
on 4th down because you can get that 50 to 75% again.

The second thing these studies don't factor in is macro-game situations.  Like if you're
up by 2 late in the game, maybe if you go for it and get it then you greatly increase
your chance to get 7 more points and winning by 9, where if you fail then you greatly
increase you chance of losing by 1.  This is similar to all the anti-bunt numbers in
baseball.  All of them use run probability to conclude that giving up the out lowers
your chance at a big inning.  But in a lot of scenarios late in the game you don't need
a 7 run inning, you just need that 1 run, and you'll trade away the chance for a big
inning for a chance for 1.

Sure, some situations might call for more aggression on 4th down, but there's way too
many interconnected aspects of this to devolve it to a simple 'solve for x' kind of problem.

edit - Also, they don't factor in individual team's abilities.  If you have a bad ass defense and
a horrific o-line, you're more likely to benefit from kicking/playing conservative on offense vs
if you have a fantastic o-line and bad defense.

Well, yeah... As far as the football decisions involved here go, these are all baseline, average calculations. He's looking at a league-wide sample across 3 seasons and focusing only on fourth down situations in the first quarters of games.

If you were to try to apply this study practically on the field, you'd absolutely need to take into account game-, team- and situation-specific factors. And with football, far more than a game like baseball, strategic decisions on the part of the play callers is a huge factor, too.

He accounts for some of these variables in various ways, in part by simply bracketing them out:

By describing the values of situations in terms of expected point differences, I am implicitly assuming that a team that wants to maximize its chances of winning should be risk-neutral over points scored. Although this is clearly not a good assumption late in a game, I show in Section IV that it is an excellent approximation for the early part. For that reason, I focus on the first quarter.

Focusing on the first quarter has a second advantage: it makes it reasonable to neglect effects involving the end of a half. Because play in the second quarter begins at the point where the first quarter ended, the value of a given situation in the first quarter almost certainly does not vary greatly with the time remaining.

But, again, Romer's purpose was not to draft a strategy guide for NFL football.

This was a paper on economics and the assumption that rational agents maximize choices.

(And I think the disparities between observed behavior and his analysis' recommendation of optimal behavior is big enough that his overall economic conclusions—that firms don't necessarily maximize outcomes the way simpler models assume, for reasons that may be unclear—probably remain valid or at least relevant, even if his baseline analysis is off the mark in many specific situations.)

[TDubbs]

The problem with these kinds of studies is that they ignore the fact that 4th down
plays don't exist in a vacuum.  First off, much of NFL defensive strategy is predicated
on the fact that if you hold a team on 3rd down they are going to kick, this
frequently inflates the yardage on 3rd down when defenses give up the 4 yard check
down on 3rd and 8 or whatever.  In this genius' math, he uses 3rd down results to
predict 4th down results.  So the fact that defenses will usually let a team pick up
50 to 75% of 3rd and long influences his math to think that of course you go for it
on 4th down because you can get that 50 to 75% again.

The second thing these studies don't factor in is macro-game situations.  Like if you're
up by 2 late in the game, maybe if you go for it and get it then you greatly increase
your chance to get 7 more points and winning by 9, where if you fail then you greatly
increase you chance of losing by 1.  This is similar to all the anti-bunt numbers in
baseball.  All of them use run probability to conclude that giving up the out lowers
your chance at a big inning.  But in a lot of scenarios late in the game you don't need
a 7 run inning, you just need that 1 run, and you'll trade away the chance for a big
inning for a chance for 1.

Sure, some situations might call for more aggression on 4th down, but there's way too
many interconnected aspects of this to devolve it to a simple 'solve for x' kind of problem.

edit - Also, they don't factor in individual team's abilities.  If you have a bad ass defense and
a horrific o-line, you're more likely to benefit from kicking/playing conservative on offense vs
if you have a fantastic o-line and bad defense.

Well, yeah... As far as the football decisions involved here go, these are all baseline, average calculations. He's looking at a league-wide sample across 3 seasons and focusing only on fourth down situations in the first quarters of games.

If you were to try to apply this study practically on the field, you'd absolutely need to take into account game-, team- and situation-specific factors. And with football, far more than a game like baseball, strategic decisions on the part of the play callers is a huge factor, too.

He accounts for these variables in various ways, in part by simply bracketing some of them out:

By describing the values of situations in terms of expected point differences, I am implicitly assuming that a team that wants to maximize its chances of winning should be risk-neutral over points scored. Although this is clearly not a good assumption late in a game, I show in Section IV that it is an excellent approximation for the early part. For that reason, I focus on the first quarter.

Focusing on the first quarter has a second advantage: it makes it reasonable to neglect effects involving the end of a half. Because play in the second quarter begins at the point where the first quarter ended, the value of a given situation in the first quarter almost certainly does not vary greatly with the time remaining.

But, again, Romer's purpose was not to draft a strategy guide for NFL football.

This was a paper on economics and the assumption that rational agents maximize choices.

(And I think the disparities between observed behavior and his analysis' recommendation of optimal behavior is big enough that his overall economic conclusions—that firms don't necessarily maximize outcomes the way simpler models assume, for reasons that may be unclear—probably remain valid or at least relevant, even if his baseline analysis is off the mark in many specific situations.)

But where does the meat go?

[Quality Start Machine]

I'm guessing Romer's paper came out of a masterplan in which he got grant money and used it to go to football games.

[Kermit, B.]

I'm guessing Romer's paper came out of a masterplan in which he got grant money and used it to go to football games.

I disagree.  He probably submitted it as a thesis and earned himself some worthless college degree.

[MAD]

I'm guessing Romer's paper came out of a masterplan in which he got grant money and used it to go to football games.

Who do you think he is, Weebs?

[Quality Start Machine]

I'm guessing Romer's paper came out of a masterplan in which he got grant money and used it to go to football games.

Who do you think he is, Weebs?

No, because it didn't ruin football for anyone.

[CubFaninHydePark]

Wow. This was a bit harsh.

I blame falling asleep on the couch and waking up to Andy Dick on my TV for putting me in a foul mood.

Here's the argument against this in statfag terms...

http://wp.advancednflstats.com/winprobcalc1.php

With 4:08 remaining in the 2nd quarter...

First and 10 on your opponent's 13:

Expected Points: +4.5
First Down Prob: 0.67
TD Prob: 0.54
FG Prob: 0.33

First and goal on your opponent's 9:

Expected Points: +4.8
First Down Prob: 0.00
TD Prob: 0.60
FG Prob: 0.28

I'm not sure if this is due to rounding or not, but you left out the ultimate statistic that supports me: Win probability.

In the two scenarios:

1st and Goal at the 9: Win Probability:    0.25

1st and Ten at the 13: Win Probability:    0.26

What I'm not clear on is whether the TD probability is your odds of scoring a TD at any point in the future or on that series.  Given that it breaks down 1st down and TD probability, I could see it being based on that series, not inclusive of other series.  I think that this is right:

If you have a .67 chance of a first down, but a .54% chance of a TD, that leaves 13% of the times you're getting a first down without scoring.

If you then have First and goal from the 3 (worst case 1st down scenario--it only gets better), you have a 78% chance of scoring a TD.  When you take that times the 13% of instances from the 13 you're getting a first down without scoring, you get an extra 10% chance of scoring a TD.

So your TD probability (maybe) from your own 13 is really .54 + .10 = .64.

That .04 difference is the only thing that I can think of to explain the win% discrepancy.

But if you want to get all statfaggy with it, at least include the probability that truly matters, even if it (ever so marginally) doesn't support your argument.

[Dr. Nguyen Van Falk]

What I'm not clear on is whether the TD probability is your odds of scoring a TD at any point in the future or on that series.  Given that it breaks down 1st down and TD probability, I could see it being based on that series, not inclusive of other series.  But I also could see it being the other way.

I'm almost certain it's the odds of that drive going forward resulting in a touchdown.

That is: on average, drives end in touchdowns 60% of the time when an NFL team has to proceed from first and goal at its opponent's 9, and 54% of the time when it proceeds from first and 10 at its opponent's 13.

But it's interesting that the scenario w/ the lower expected value of points has a higher overall win%.  I'm not sure how to explain it, other than it's not including probabilities of a TD after you get a first down.

These probabilities are based on actual game data stretching back to 2000. It wouldn't shock me at all for there to be a few slightly unsmooth spots in the probability curves.

And the 1% difference is quite marginal, particularly when what you're talking about is the probability of winning/losing the entire game, as opposed to the probability of scoring on a single drive.

But if you want to get all statfaggy with it, at least include the probability that truly matters, even if it (ever so marginally) doesn't support your argument.

The win probabilities take into account actual game situations (score, time, and so forth) and reflect probable game outcomes, and thus reflect decisions and plays made later in the games in the sample (in this case, including the entire second half).

The scoring probabilities are generic (ignoring the clock and scoreboard), and reflect only (as far as I can tell) probable drive outcomes, regardless of situation.

Considering the essence of your argument more or less boiled down to "a first down at your opponent's 13 yard line is, generally speaking, better than a first down at their 9 yard line" (as opposed to, say, "a first down at your opponent's 13 is better than a first down at their 9 when you're down by 13 with around 4 minutes remaining in the first half"), I chose the generic probability.

Which probability suggests what most of us intuit: that getting closer to an opponent's goal line makes it more likely that you'll score a touchdown on that drive.

[Dave B]

Thanks to holding penalties and dead-ball unsportsmanlike and personal foul penalties, we had a game this year where a team went from first-and-goal at the nine to first-and-goal from their own 41 yard line in two plays. I wonder what the odds became at that point.

[CubFaninHydePark]

The scoring probabilities are generic (ignoring the clock and scoreboard), and reflect only (as far as I can tell) probable drive outcomes, regardless of situation.

Considering the essence of your argument more or less boiled down to "a first down at your opponent's 13 yard line is, generally speaking, better than a first down at their 9 yard line" (as opposed to, say, "a first down at your opponent's 13 is better than a first down at their 9 when you're down by 13 with around 4 minutes remaining in the first half"), I chose the generic probability.

Which probability suggests what most of us intuit: that getting closer to an opponent's goal line makes it more likely that you'll score a touchdown on that drive.

Fair enough--the win probability could be noise, but it is a large sample size, which makes me think that it still could suggest something.  I wonder if teams in 1st and long-goal situations are more prone to go for it on 4th down, which might go a long ways toward explaining the TD/FG differential (that almost lines up perfectly).

[Slaky]

Let's simplify:

YOU WANT TO GET CLOSER TO THE GOAL LINE AND SCORE. YOU PLAY TO WIN THE GAME. HELLO.

?