No One Remembers Shockdome
Originally intended as a UV bit, but then it got long.
As you've probably heard, Bill Belichick went for a first down on fourth and two from his 29 with about two minutes left and his team leading by six points. The Patriots didn't get it, the Colts made the short march for the game-winning score, and commentators duly exploded at how awful the decision was. Tony Dungy evidently kept saying Belichick "should have gone with the percentages."
He did. Of course he did, he's Bill Belichick:
With 2:00 left and the Colts with only one timeout, a successful conversion wins the game for all practical purposes. A 4th and 2 conversion would be successful 60% of the time. Historically, in a situation with 2:00 left and needing a TD to either win or tie, teams get the TD 53% of the time from that field position. The total WP for the 4th down conversion attempt would be:
(0.60 * 1) + (0.40 * (1-0.53)) = 0.79 WPA punt from the 28 typically nets 38 yards, starting the Colts at their own 34. Teams historically get the TD 30% of the time in that situation. So the punt gives the Pats about a 0.70 WP.
This is obvious in retrospect, right? Bill Belichick is one of the greatest coaches in the history of the NFL. There is a slight chance he knows what he's doing. And even if you are one of the folk who really believes emotion and momentum overwhelm probability in football, not even broaching the idea that Belichick might be on to something is simultaneously stupid and arrogant—neat trick. If you are a caveman when it comes to football and see Belichick go for it there, your first thought should be "hmmm… maybe I don't know something." This, obviously, has not happened. Caveman status is self-perpetuating.
Big Ten Wonk's civilian alter-ego notes that Joe Posnanski may have disagreed with the decision but at least showed Belichick—and math—respect by mentioning the percentages:
Really, no matter how you play with the numbers, it will come out about the same. Try it. There is almost no way–without suppressing the numbers–to make the percentages even out. The Patriots’ best PERCENTAGE chance was to go for it on fourth down. Of course, football is not really a percentage game for most of us, is it? No, it’s a game about emotion and passion and momentum.
This is where Posnanski needs to play a lot of poker. Emotion and passion and momentum are great for football players. For coaches they are ways to go on tilt and make dumb decisions that are safe but go against the percentages. Belichick is ruthless and in a position where media criticism means nothing as far as his job goes. Most other NFL coaches would take the safe route and decrease their chances of winning because they perceive that it would increase their chances of keeping their job.
But, you know, at least Posnanski brought it up. Many thousands didn't, and just did some blah blah about how it was dumb, thereby implying they were smarter than Bill Belichick. This is why he is the traditional journalist who's bridged the internet divide more successfully than any other. Clark Wonk:
In other words, ”traditional sports punditry” is denoted not by what kind of resume you have, how old you are, whether you sit in the press box, or even whether your thoughts are packaged in 800 words of ink, 1600 words of pixels, or two minutes of streaming video. No, “traditional sports punditry” denotes merely that you’re not staying current within your own field: “What the hell is Belichick doing?” as opposed to “Whoa, talk about trusting the percentages–what the hell is Belichick doing?” …
To be aware of what Posnanski calls the “PERCENTAGE”s, ones that indicate that probability was in Belichick’s favor over the course of a thousand tries, does not rule out disagreeing with the coach in this single instance. But to not be aware of these percentages is to fail in the most basic journalistic sense. To write about a decision, much less try to criticize it, without displaying any understanding of its self-evident context is to fall down on the job in the ”why” department, even if you do get the who, the what, the when, and the where.
Humans would be well-advised to nail the “why,” by the way. Computers can now do those other four pretty well.
Seriously. Does anyone remember SHOCKDOME XXXVI*? The Patriots get the ball back in a tie game with about two minutes left. Tom Brady is the Patriot's first-year starting quarterback. John Madden, embodiment of conventional wisdom-type substance, is publicly begging Belichick to run the clock out and head to overtime instead of putting the game on the kid's shoulders. Belichick says screw that noise, I've seen Tom Brady play football, let's go, and two minutes later Adam Vinateri is kicking a game-winning field goal and no one remembers the ballsy decision that won the damn game. The reason Belichick is so untouchable that he can defy conventional wisdom is because he defied conventional wisdom.
And yet no one mentions this.
*(AKA Super Bowl XXXVI. An aging Pat Summerall awkwardly blurted out "this game has turned into Shockdome 36" at some point in the second quarter when the Patriots weren't getting stomped like everyone expected they would, and I died laughing.)
I heard the talk radio bloviators bloviate and yammer on endlessly about what a terrible, indefensible call it was blah blah blah yada yada yada. Even without resorting to stat geek numbers the decision was the right one. NE's defense was exhausted, having just been shredded by Manning to the tune of a 70-yard-plus TD drive consummated in just under 2 minutes. His defense was also wracked by injuries to the point that Harrison and Bruschi would have been upgrades had they suited up. I mean, they were playing Leigh Bodden, a guy who wasn't considered good enough BY THE LIONS to keep around. Belichick had a good offense that moved the football well throughout the game, an outstanding QB in Brady, and if they get 2 yards they win the game. Forget about the position on the field; if you can get 2 yards and win, you should do that. Besides that, they had the whole 'are they going to go for it or just try to draw the defense offsides' thing going for them, meaning they should have had a slight initiative advantage as the defense makes sure to wait and see the ball snapped. Belichick should do the exact same thing in a similar situation.
I am pleased that people who KNOW football (Brian, Chris from Smart Football, Easterbrook, stat geeks everywhere), as opposed to people who talk about football, support the call. Conventional wisdom often isn't, and common sense DEFINITELY isn't.
The problem with basing the 4th down decision on historical data is that it none of the historical data takes into account 4th and 2 from your own 28 with a 6 point lead and 2 min left. Unless this situation has been played out before which I seriously doubt.
In football little details change the strategy significantly. It would be interesting to see, historically, how many of those 4th down attempts where runs and passes and the conversion rates of each.
I think Belichick is a great coach who probably made a bad decision. Like a great coach, I doubt he evaluated the aggregate generalized historical data as much as he focused on the factors involving his specific situation. Many of the factors we fans watching the game on TV know little about.
I also heard someone on SiriusXM radio (it was on the Mad Dog channel, but wasn't Dog himself) actually say that a reason to punt there was so Belichick wouldn't have to listen to the press second-guess him if they didn't get it - so a coach should make his decisions based on the media perceptions?
Although, I wonder if this is a reason more coaches don't do stuff like this - going against "conventional wisdom" can be a problem if your job's at stake. (I think there was a post a month or so ago all about the punt/go-for-it decision bringing up this same point).
Jeff
"But, you know, at least Posnanski brought it up. Many thousands didn't, and just did some blah blah about how it was dumb, thereby implying they were smarter than Bill Belichick."
And here's the worst I've seen of the thousands who didn't (pay heed to percentages), coming from MGoBlog's new favorite punching-bag, the one-and-only Michael Wilbon:
http://views.washingtonpost.com/world-wide-wilbon/wilbon/2009/11/belichi...
Wilbon, a sportswriter with no coaching or psychiatric experience, has the arrogance to identify a character flaw in Belichick and cite it as the purpose for him losing a game?
I love how he makes going for it on fourth down a huge unknown, but punting a sure thing. In Wilbon's mind, the worst thing that can happen going for it is a fumble, or the QB trips, things like that. But if they punt, the worse thing that can happen is a 70-yard drive. Really? Punts never get blocked? Snapped over the punter's head? Shanked? Returned for TDs?
don't account for is any kind of human element, regardless of the W or L. Whether BB was playing the percentages or not, the team's perception of the call is something he's having do deal with this week (along with an opinion about the call from every corner of the earth, obviously). Maybe he's good enough to convince his defense that he has all the confidence in the world in them, but that's going to be a tough sell. From my point of view, I like the contrarian call, but it wasn't made in a mathematical vacuum where the only outcomes were A) first down and win or B) turn over on downs and lose. Had he punted, the Colts very well may have marched the field and still won, but the Pats would be dealing with fewer distractions this week.
Reason #umpty-zillion why I love MGoBlog:
makes me think of my favorite paper in Psychology: "Unskilled and Unaware of It: How Difficulties in Recognizing One's Own Incompetence Lead to Inflated Self-Assessments:"
http://www.apa.org/journals/features/psp7761121.pdf
The 2005 season convinced me that sometimes you have take risks on offense to try to keep your defense off the field...I remember OSU getting the ball on the last drive of that game and being completely sure that they'd score.
"That's horrible"
"Terrible is what it is."
The New York Times has a story about the Zeus Computer Program this morning and how it comes out the same way as most of the posters in this thread have indicated: http://fifthdown.blogs.nytimes.com/2009/11/16/zeus-computer-program-supp...
I think these are a bit misleading. Assuming I didn't miss something here, its a 79% win probability against an average team. Pey Pey, Reggie Wayne, and Dallas Clark are not average. You have to figure this increases the average WP of starting from your own 34 above 70% and increases the WP of starting at the Pats 28 above 53% (and thus decreasing the 79% WP). This isn't to say that punting will necessarily be the mathematical choice, but I think using average offensive statistics here doesn't quite tell the whole story.
If you followed the link, you'd see that they addressed this point in the post and determined that, while relevant, it doesn't change the outcome: the better choice was to go for it.
http://www.advancednflstats.com/2009/11/belichicks-4th-down-decision-vs-...
by not addressing it. There's some fancy hand-waving without showing us the analysis. I mean sure if you assume there is a constant multiplier for the probability of scoring from each field position because they're the Colts and not the Rams, of course its not going to do anything (for example, the Colts are 125% as likely to score as the average team regardless of field position). It probably even makes the decision better. But I think the closer you are to the endzone, the more you have to consider the quality of the offense which tosses a "team multiplier" out the window. I'm not saying that it changes the outcome, I'd just like to their particular reason for dismissing it.
I'm not sure that's the assumption here (although, if it is, you're right). I think they're just stating the obvious fact that the Colts offense is better than the NFL average, so any analysis that takes this into account only strengthens the argument to go for it rather than punt (i.e. they are more likely than the average team to score from any particular position on the field - though this may or may not be some constant multiple over the entire domain of the density function). It's really more or a hand-wavey qualitative argument than quantitative argument, but the logic holds.
Now, you might be able to make a similar argument based on the relative quality of the Pats' defense compared to the league average, but (and this is highly hand wavey) I would wager that the quality of defenses is normally(-ish) distributed, and the deviation isn't that high. Just vaguely, I'd guess this effect isn't that large (though, obviously, in lieu of actual data this isn't a very convincing argument). The point being, without having a defense that is substantially below the league mean, this probably doesn't change the outcome.
And I'm satisfied. Fun game that football...
if you are a reporter, you just write one column the day after with your emotional reaction that Bellicheat is a fool. Then two days later you turn around and write another column, pinching someone else's actual journalism about how it was a great decision and everyone needs to back off.
Twice the columns, less than 1/2 the work.
It only took me 25 years to get smart and head to SD
is why is this one coaching decision and play is being beaten to death with a shovel by ESPN Sportscenter and all sports networks? Come on! It's one game. Had it worked, the game and all this ridiculous flailing about would be ovah! Oh, and we can be reasonably sure the Pat's season is far from over.
Besides, that line judge needs an eye exam. The mark was crap.
Markusr2007
http://www.maizenbrew.com/
Is the Shockdome anything like the Terrordome?
http://www.youtube.com/watch?v=9WIeEuRtEW8
but MGoBrian, I have a follow up question.
You're big on these in-game probabilities, but dont seem to buy into historical probabilities, like when I pointed out how poor ATS Michigan is in their first road game. or how great MSU is in their final home game of the year?
Whats the differnce, IYHE?
Both are probabilities with overwhelming percentages one way. You're asking to lose if you bet on UM in their first road game.......just like you're increasing your chances of losing by punting in that spot on 4rth and 2.
The Winter Olympics at www.justcoverblog.com
God Bless Your Cotton Pickin' Maize & Blue Hearts
Season based probabilities are poorly controlled and poor predictors. You don't have more than a sample size of at most 100. Position and time based probablities seem to control much better for the important factors.
True, but MIIICH's struggles ATS in their first road game stretch back over 30 years. It's part of the program's ID at this point, dont you think.
Same thing with MSU's 19-3 ATS record in their final home game of the year. That's a regular occurence that's happened for more than 2 decades over the span of a 5 different head coaches.
I hear you on small sample size and lack of control, but there are certain historical trends out there that cant be overlooked, IMHE
The Winter Olympics at www.justcoverblog.com
God Bless Your Cotton Pickin' Maize & Blue Hearts
There's another factor here besides sample size, and that's sample composition.
If Michigan has a bad year or a run of bad years, there may be extenuating circumstances that led to those results (like, say, years of lousy recruiting or a coaching change). Because of the sample size, those circumstances can have a substantial effect. If we go back to your example of the last 30 years and we agree to chalk the last two years up to, say, black magic, that's 6.67% of your sample size that isn't "average." That's a significant amount.
In game probabilities, on the other hand, have colossal sample sizes that encompass most any combination of team skill level, outside circumstances, coaching ability, weather, or karma imaginable. Somebody made an argument that the Colts offense is better than average; that's irrelevant, because the "average" in those percentages includes teams as good as the Colts, better than, and far, far worse.
For the record guys, I dont disagree with Belichek's move or the probability analysis behind this post and the defense of him going for it.
I just think you probability guys dont lend enouigh credence to the power of program and college football history when it comes to some of the ATS angles/trends.
When looking at ATS trends you have to skeptic first. It takes a time to figure out what is a valid trend or not.
The two I listed are. When somebody says team A is 6-1 ATS on Tuesdays is not, for example.
The Winter Olympics at www.justcoverblog.com
God Bless Your Cotton Pickin' Maize & Blue Hearts
I actually would like to have seen the Patriots go for their previous 4th down on their drive before...instead of kicking the Field Goal.
They were up 10 points and I would rather try and go up 3 scores (17 points) than kick a FG and remain in a situation where 2 Colts TDs costs you the game.
With that in mind, I think Belichick should have ran that second last series once they got in Field Goal range as a series willing to use all 4 downs to put it in the endzone and go up 3 scores.
for other points in the game. If its a good decision to go for it with 2:00 min left, is it also a good decision to go for it in the 1st or 3rd quarter?
I happened to love the decision. Most coaches err on the side of conservatism and people think they are risky b/c they run a triple reverse at some odd time.
It only took me 25 years to get smart and head to SD
with a lead in the 3rd quarter this season. They went for it on fourth-and-1 from their own 24 and made it. The Patriots went on to control the ball and the clock for the rest of the quarter and put the game away.
Probably not. The statistics are easier in this case, because a 4th and 2 conversion effectively wins the game. In the 1st or 3rd quarter, it's just maintaining possession a long way from the end zone. I don't think there is enough of a reward for converting to balance the risk of failing.
Our helmets got wings
How does this decision making process factor into Iowa's play for overtime at the end of the OSU game?
While watching I was thinking the probabilities of going 50 yards for a field goal were much higher than going 25 yards for a score and having to play defense as well.
In fact, because Tressel had already shown he was playing for OT, there was virtually no downside to going for the win. It was playing with house money, and Ferentz folded.
While I didn't agree w/ Ferentz playing for OT there, I could see why he did it. He had a QB making his 1st start playing in Columbus. A turnover there gives OSU a short field to kick a winning FG. That said, the kid was playing quite well considering the circumstances.
I would love to know the statistics on the outcomes when teams play for OT like that. I know I was in the minority, but I wanted to see RR go for 2 against MSU and get the win right there. I thought we would be playing at a disadvantage in OT since we were on the road and MD could no longer sit back and coach not to lose.
That qb making his first start had already managed to bring them back to tie the game. The decision to play it safe and go to overtime seemed to be based on Ferentz's unfounded fear, not on the qb's performance to that point.
sgtwolverine.com
Iowa's decision that trying to win the game wasn't a good path brought back many frustrating Michigan memories from the past decade.
This example illustrates how Belichick uses stats to make his game decisions, but his wisdom goes far beyond that.
In draft-day decisions, notice how NE is always trading high picks for lower ones (eg taking a chance on longshots like neer-do-wells like Brady, although he's not actually the result of a trade). Also, notice how NE often trades current year picks for future ones (this year NE also traded one of his top defensive players, Seymour, for Oakland's #1 pick 2 years hence, when there will probably be a rookie salary cap).
Why do these strategies make sense? First, there's evidence that it's more cost effective to trade away high picks (who, based on past performance statistics, cost more than they're worth). Since a team has a fixed budget, the best teams are built by making cost-effective decisions--getting the most bang for the buck. Second, the best teams are built by making cost-effective decisions for the future and resisting the pull of one's desires to get the best players immediately. Other teams often choose the immediate reward of getting (or keeping) a player now in lieu of the delayed reward of getting an even better (and/or cheaper) player later. As an economist, Belichick realizes this bias. And he exploits it.
PS On this blog, yesterday, I published a decision analysis which demonstrated Belichick's wisdom in the IND game (dakotapalm pointed out that someone else already also laid out the logic for the decision). But it was nice to see that some objective numbers from NFL stats actually reached the same conclusion.
http://mgoblog.com/mgoboard/ot-4th-and-2-your-own-territory-near-end-game
michelin
Does this make Notre Dame less likely to offer him their head coaching job (which he'd clearly take)?
What makes you think that? Just curious...
I believe you need to recalibrate your sarcasm meter
You can't arbitrarily just accept 60% chance of making 4th and 2. How big is this sample size, what's the confidence interval, etc. An average alone is not necessarily statistically significant. Moreover, (and this is from a comment I read elsewhere), a more telling stat is 2 pt conversion success, not the entire population of all 4th and 2 chances ever. This percentage is, in fact, lower. You can't always simply things like this down to a decision tree like you did, you need to question the validity of the stats you are using to make your argument. There are a million factors in play here and most aren't taken into account here. Also you can't just assume that Belichek is risk-neutral, which you also do.
difference between a 2-point conversion and 4th and 2 from midfield. There's a reason a lot of teams get into two TEs and pound the ball from the 2--the shortened end zone plays a huge factor into what you can do successfully.
I will agree that there a lot more factors in play than the analysis above. My guess is that the decision is pretty close to a wash and Belichick put the game in the hands of his best player to make a play.
Actually, I think the two-point conversion point is a really good one: in this case, Indy does not care about a 60-yard play. Two yards is the same as 60 to them: a loss. So they can play it like it's fourth and goal from the two.
It's still considerably easier to get two yards than three, so the conversion rate should be well above 50%.
MGoBlog | Sporting News | HTTV 2009 | e
the NFL does their 2-point conversions from the 2, not the 3
I've read elsewhere - on Advanced NFL Stats, I believe - that the two point conversion percentage is weighed down by one-point conversions that go bad and no kick is attempted. It doesn't happen often (except to Romo), but neither do two-point conversions.
I also agree that defending the entire field makes the chance of conversion higher than defending 12 yards. Game situation is irrelevant; you still have to stop them from getting 2 yards... or 60.
Indy is going to let Randy Moss go down the field unimpeded. They're going to cover him if he goes past 15 yards. They aren't going to cover him if he goes into the tunnel on the goalline. I agree that Indy needs to protect the sticks like it's the goalline, but NE has a lot more options to work with in the extended field.
If the probability of winning is higher by going for it rather than by punting, then it's a no-brainer.
michelin
Why would 2pt conversion success be more telling? 2pt conversions have the disadvantage of working w/ a shorter field making the conversion more difficult.
...because I have no idea as to the accuracy of that statistic, but I disagree that 2 point conversion success would be more telling than taking the success rate of the entire population of all 4th and 2 chances ever. 2 pt conversions are going to be notably more difficult than your "average" 4th and 2 plays because on a 2pt conversion the defense only has 12 yards to defend, making it more difficult to get those two yards than it would be if they had to defend the "whole"* field.
*Allowing for the fact that the Colts in that situation wouldn't be ACTUALLY defending the whole field, but would almost certainly be forced to defend more of it than just 12 yards.
Why is the 2-point conversion "more telling" in this case? There are a lot of reasons a 2-pt conversion would be a lower percentage conversion than a 4th-and-2 from the 28 yard line (think amount of field to be covered by the defense).
Brewer. Wolverine.
It's certainly not the same as a 2-point conversion, but I would argue that it is closer to a 2-point conversion than your average 4th and 2.
I am willing to bet that a very small percentage of 4th down plays occur when a team does not have the ultimate goal of scoring a touchdown. Most come when the team is down and needs to score or comes at a position on the field where the risk of not making it does not outweigh the field position benefit of punting. In both these circumstances, a defense is forced to play it just like a 3rd and 2. Know that the offense is likely to attempt a short yardage gainer play, but have to defend against the touchdown. The safeties still can't get beat deep and the cb's can't abandon deep coverage to cheat on the five yard routes.
In this situation, as Brian says below, if NE gets 3 yards or 60, the game is over. Indy can sell out to defend the plays that from the offensive perspective are most likely to achieve at least 2+ yards. From a realistic standpoint, they are defending the run and passes within 15 or so yards of the line of scrimmage. Yes, they can't ignore Moss streaking down the field, but they also don't have to pay as much attention, sinceif they get beat initially and have to scramble that's a much tougher pass and probably not worth the gamble for Brady. On both sides of the ball, they treated the play as a two point conversion, and that is the more relevant statistical measure.
The only thing that really is a big advantage for the fourth and 2 call as compared to the two point conversion is that when the QB scrambles, receivers are not confined by the back of the endzone, tilting the odds of completeion in favor of the offense.
If I had to judge based on the two statistical measures, I'd say it's somewhere between the success rates of two point conversions and regular 4th and 2 situations, but closer to the two point conversion success rate.
"If you are a caveman when it comes to football and see Belichick go for it there, your first thought should be "hmmm… maybe I don't know something." This, obviously, has not happened. Caveman status is self-perpetuating."
Brilliant.
And to Tedy Bruschi, if your defense is so insulted, shouldn't you be MORE insulted that you let Manning march down the field twice in the fourth quarter already? If you think you are good enough to stop them from 60 yards, okay, stop them from thirty yards. This is what happens when football commentators respond from emotion instead of from fact.
Rodney Harrison, another exPAT, also went ballistic. I am sure that Belichick finds their antics amusing. He realizes he didn't hire these guys for their wisdom.
michelin
the only thing more uncomfortable than Tedy Bruschi screaming at the ESPN camera, in studio, was Chris Berman who went on telling him afterwards about how they could use the showers together b/c that's what everyone at espn does. Anyway, I was worried that Bruschi was going to have another stroke, or a heart attack or something. Someone at ESPN needs to calm him down.
i only respect other superfans
http://mgoblog.com/mgoboard/ot-neindy
Victory is mine!
It is hard to believe he is from Ohio.