Comments on Advanced NFL Stats: Worst 4th Down Decision of 2008

Is there a column on worst 4th down call in 2009? ...

2010-06-05T22:35:51.764-04:00

Is there a column on worst 4th down call in 2009? Because the Bears went for it on 4th and 11 from their own territory in the GB opener, Cutler's 1st game. And there was no apology. Their long snapper "had the green light" to call it if he saw a certain formation. Only the formation changed right when he put his head down to snap. Inexcusable coaching on this one.

I'm not sure. If you're talking calendar ...

2010-06-05T22:33:37.360-04:00

I'm not sure. If you're talking calendar year 2008, Bill Belichick went for it on 4th and 13 from the Giants 31 in SB XLII. Yes you read that right. In the 3rd quarter of a game he lost by 3 points.

Well before the more recent 4th down controversies, Belichick was making "genius" moves like this one. It's comic, but no one remembers since the 4th quarter was so great. This set it up though. Let's just give up 3 points there [or expected 2.2pts let's call it)! Sure it was Gostkowski not Vinatieri but seriously.

-Brian I'd love to see you test "the mom...

2009-09-11T04:31:15.439-04:00

-Brian

I'd love to see you test "the momentum hypothesis." It seems if the momentum hypothesis is true then in situations where a team goes for a 4th down conversion and is not successful, the standard win probability model given by down, distance and time left should be higher than the actual observed win probability.

For example, fictional team in Philly's position goes for 4th and 1 at 49 and fails. They should win less than the amount predicted by your WP model. The null hypothesis is that teams are modeled the same way whether they just failed to convert a 4th down or not. The alternative is to reject, teams are worse than a similar team in the same position after failing to convert on 4th.

I would bet against the momentum effect, but a thorough analysis would be interesting.

Charlie

Bruddog, There is a much more complicated calcula...

2009-09-10T17:27:03.315-04:00

Bruddog,

There is a much more complicated calculation that you can do that would take into account the relative strengths of the team but it actually won't change the calculus all that much.

For one thing as Brian said, overtime is most of the time based on one or two possesions and a sample size that small will lessen the talent element and increase the luck element, especially when the single largest variable (the coin) is pure luck.

For another thing, the calculations can cut both ways.

Take extreme scenarios:

PHI has the best defense known to man.

Phi's defense will probably stop them even if they have the ball at the 50. Its not that big a risk to go for it.

PHI has the worst defense known to man.

Phi defense will give up a score even if you punt. You need to take your chances with your offense.

PHI has the best offense.

A fourth conversion is highly likely. Go ahead and go.

PHI has the worst offense.

They won't score when they get the ball back later. Might as well take the risk now.

Obviously, the long calculation needs to calculate the odds either team will score from the various points on the field and each of those calculations is dependent on the talent level of the various squads. However, the calculations tend to mostly cancel each other out.

Using the league average conversion rates isn't as bad as you might think.

Bruddog-That's a good thought, except you almo...

2009-09-10T11:02:07.611-04:00

Bruddog-That's a good thought, except you almost never get a full quarter for the better team to come out on top. In fact, it's just the opposite. It's whoever scores first, and that's heavily dependent on the coin flip. It's very random.

I agree that most assessments of 'how the game...

2009-09-10T11:00:31.027-04:00

I agree that most assessments of 'how the game is currently going' are based on very small samples.

Ideally, you'd want a Baysian-type approach, where you start with the league baseline or team baseline performance. Then based on current game performance, you'd move the needle slightly one direction or the other.

However, no matter how the particulars of the matchups were going that day, the Eagles only needed a 5% success rate for the conversion attempt to make sense.

Look at it from the point of view of a Bengals fan. If the Eagles instead had called a time-out, brought off the punt team, and lined up to go for it, what would you be thinking? Are you thinking "Sweet, we'll stop this run for no gain and then get the ball in even better field position?" No. Of course not.

"Oh S#!+" is what you're thinking!

There's a good reason why that would be your reaction. You were thinking your team would be getting the ball back with a good chance to win the game, and at worst go into overtime. Now, chances are your defense will still be on the field facing a fresh 1st and 10 and probably lose.

One thing not mentioned here is the relative team ...

2009-09-10T07:43:21.058-04:00

One thing not mentioned here is the relative team strengths. Philly had been a much stronger statistical team throughout the season. It's much more understandable that Philly might want to punt to play for overtime. It makes much less sense for the Bengals to punt in that same situation.

Before the game PHI's expected win % according to this site was 88%. I can't think of the math right now but its quite possible that another full quarter of play would have given them a better expected winning % than one single play.

I think that focusing on how a team is playing in ...

2009-09-10T05:04:03.235-04:00

I think that focusing on how a team is playing in that day's game is usually a mistake. I've looked at this from a couple of statistical angles and there is a whole lot of regression to the mean going on when you try to extrapolate things from one day's performance.

We've all seen games where a team has been playing horribly for 3 quarters and then turns it on in the fourth. We've also seen games where a team continues to suck in the fourth. You're much better off trying to project based on overall performance numbers and chaos theory than you are taking that day's performance so far and extrapolating. Basically the fact that a team is red hot/ice cold means a lot less than you think it does.

The other weird thing about the "coach looked and knew that long term it would work but not this time..." argument is that coaches basically always make the same decision. Its a little too much like Lake Woebegone where all the children are above average. If the averages say to go nearly every time and the coaches are not going nearly every time because something about the individual circumstances is different, then obviously something is off and there must be some atrociously bad decisions mixed in with the couple of times they got the circumstances right.

The final thing I don't get about the "we don't want to gamble thing" is that it is still a fucking gamble. By not going, Reid was gambling on a whole bunch of things happening and not just one. He was gambling the defense would make the stop. He was gambling that the offense would get the ball back through either a coin flip or another defensive stop. He was gambling that the offense would ultimately rise above the day's futility and score once they got the ball back. What exactly made him convinced that these string of occurences was more likely than the one occurence a fourth down gamble would have required?

It wasn't like they could have left the building with a third of a win. I can understand the urge not to gamble but if you take me to a casino and show me a roulette wheel that has two thirds of the slots red and only one third black and tell me I need to make a bet one way or the other, I know which side I'd bet on.

I hate every time an NFL coach makes a ridiculousl...

2009-09-10T04:20:56.904-04:00

I hate every time an NFL coach makes a ridiculously-timid decision to punt when they should go for it, so this is NOT a defense of Reid or Lewis. But I also want to play a bit of devil's advocate here.

While I love the statistical work that you do on this site, much of it is based upon season-long averages and regression analysis. In fact, much of your work is based upon regression analysis over the span of many seasons. But isn't there a danger here of not seeing the trees for the forest?

This is what I'm getting at. Both of those teams were in that position because they had both played atrocious games up to that point. Put more succinctly, they were both sucking - big time.

So if I'm the coach and I know that the "correct" long-term, season-long or career-long play is to go for it, but I'm also looking at the horrific play of my team up to this point, I have to admit that it must be more difficult to make that call. In other words, regardless of what your long-term analysis might say, Reid and Lewis were staring at the end of a long period of ineptitude and one could argue that nothing that took place in THAT DAY'S GAME gave them any indication that they would be successful if they went for it.

I don't think there's any way to really address this statistically, because no matter how badly a team has played in a single day, there is not enough data there to do a valid statistical analysis based on only that day's game results. But sometimes a coach can look at his team/QB/star-player and just know - intrinsically KNOW - that they don't have it on that day and that he must do everything possible to just "hang on".

Professional coaches, especially NFL ones, are not...

2009-09-08T14:11:22.772-04:00

Professional coaches, especially NFL ones, are notoriously risk-averse - rather than take a calculated gamble to improve their winning chances, they would rather shift the blame/responsibility to the players, so that they (coach) get no clearcut blame for failure

Maybe so, but put yourself in Andy Reid's shoe...

2009-09-08T13:21:27.801-04:00

Maybe so, but put yourself in Andy Reid's shoes. The Eagles were bad at short yardage when the game was on the line, the Chicago game sticking in everyone's crawl. It's a big reason why they added Peters, Andrews and Weaver in the offseason. And the Eagles were playing like crap that day.

So even if the numbers scream at you "go for it, run the ball" you're not going to get Andy Reid, who is a bad in-game decision maker to begin with, to agree in that spot.

I agree the PHI power run game may have looked bel...

2009-09-08T12:00:35.775-04:00

I agree the PHI power run game may have looked below average last year, but they weren't that bad according to the numbers:

On runs on 3rd & 4th and short:
1 yd to go: 16 of 27 (59%)
2 yds to go: 4 of 5 (80%)
3 yds to go: 2 of 3 (67%)
4 yds to go: 1 of 2 (50%)

I am shocked, shocked I tell you to see that Marvi...

2009-09-08T10:57:22.837-04:00

I am shocked, shocked I tell you to see that Marvin Lewis and Andy Reid were involved in a game with terrible in-game decisions.

I don't want to defend Andy Reid on that game (hell I don't even want to think about that game) but remember that the Eagles were not getting it done at key times in short yardage last year to the point that it directly cost them a game (against Chicago), and that definitely played a part in his already poor decision making.

Although I agree with your overall point that team...

2009-09-08T10:55:26.081-04:00

Although I agree with your overall point that teams are too timid to go for it on 4th down, I remember last year the Eagles were notoriously bad on 3rd and short. Perhaps it was simply too much to ask Reid to go against tradition on 4th down at midfield late in the game when he was already under fire for a poor short yardage game.

Correction to previous post, I noticed that his wa...

2009-09-08T10:28:30.775-04:00

Correction to previous post, I noticed that his was discussing the 4th quarter and not overtime. My fault.

You note that the Eagles would not have been able ...

2009-09-08T10:26:26.887-04:00

You note that the Eagles would not have been able to respond to a successful Bengal's score--OT in the NFL is sudden death, so isn't the issue at hand not whether the Bengals would be more likely to win for the decision, but more likely to kick a field goal? As in, wouldn't the decision facing Reid be to evaluate the probability that the Bengals could score 3 points given the current field position, not how likely they would be to win/lose (which I'm assuming is based on game-long statistics)? It would seem that given that decision, it may have been more rational to punt than to go for it when you're 15 yards away from field goal range.

Perhaps the coaches realise that by expecting long...

2009-09-08T05:04:51.518-04:00

Perhaps the coaches realise that by expecting longterm probabilities to appear in the short term,they are exposing themselves to a run of short term (bad) luck and risking the sack.

Great column

2009-09-08T04:39:33.369-04:00

Great column