Thursday, May 27, 2010

The 4th Down Consultant Weighs in......

One of the finest blogs in all the vast blogosphere is Mgoblog, Brian Cook's homage to all things University of Michigan. It is as well written & comprehensive as any blog anywhere. Cook, even originated the college football "blogpoll", the alternative answer to the AP & coach's polls. Suffice it to say the Mgoblog is on the cutting edge. One of our favorite features over there is the Mathlete, who posts on math & sports & how the two intertwine. In last week's post, the Mathlete took a look at 4th down decisions & we were immediately intrigued, particularly as no team goes for it more often than Tech. Being that we at The Barrel have an actual 4th Down Consultant, we turned things over to him for a little analysis:


A couple of quick rules of thumb:

  1. Don’t punt on the opponent’s side of the field.

  2. Really consider going for it on 4th down after crossing your own 40.

  3. Field goals only make sense if there are more than 5 yards to go and you are between the 10 and 30 yard lines. If you’re in opponent territory and these two criteria aren’t true, you should be going for it.

The above statements may be attributed to The Mathlete, of the MGoblog, an all-things Michigan blog & one of the best in the blogosphere. He recently wrote a nice piece regarding when to go for it on 4th down. It really is a nice piece, and his model is sharp and elegant. But…

As my handle here at The Barrel is 4th Down Consultant, I feel obliged to respond. As always, I make the following disclaimer:

I believe that coaching staffs at all levels of football punt far too often, often out of fear. See mountains of evidence for the NFL in Gregg Easterbrook’s TMQ column for more detail. I believe this is number 16 out of 1,001 reasons why Chan Gailey sucks donkey balls.

However, I stop short of demanding that our teams go for it on some prescribed frequency; be it every time, most of the time, or 23.2% of the time when playing on a neutral field against a Cover 2 in the 3rd quarter if your yards per completion is averaging better than 6 at mid-field on the 4th play of drives starting outside the 20 when the temperature is between 66oF and 74oF.

I received the 4DC moniker after repeatedly questioning CPJ’s play calls regarding this subject in his inaugural season (undeservedly so, I think, but I wear it with pride). I still believe that he made a number of questionable calls in two games in particular (and I still think the onside kick to start the 3rd quarter against that Thursday night Miami game in 2008 was unnecessary), but I fully acknowledge that his methods and strategies over the long haul work, and its obvious that as our offense has gotten better, our ability to make 4th downs has improved. I also believe that CPJ’s style of football lends itself to 4th down plays, because while the plays we run are intricate, the play-calling is quite simple and the need for “tricks” is lacking.

That doesn’t mean that I still have to agree with each critical down decision. If I disagree with a 4th down call 10% of the time, then I’m still wondering “what the fuck” once per game when Tech plays. Nobody is above scrutiny in my book, not even the greatest college coach of our time. That doesn’t mean I think I can do it better than him. I most assuredly can’t. It doesn’t even mean I’m right. Nobody is right in “woulda coulda shouda” situations, because 20/20 hindsight is not a testable parameter.

But everybody agrees that 4th down decisions can’t be made out of emotions, be it confidence or fear. In critical situations, critical thinking is of the utmost importance. On this point, the Mathlete and I agree.

The question then becomes what kind of critical thinking? Statistical analysis is an important tool for almost everything, from medicine to football. But is there any room for qualitative analyses?

Baseball is game of numbers. Football is a game of inches. The reason is because baseball has very few working parts. Pitch, hit, run, catch and throw. Using nothing but those variables (and time, measured either in outs or innings) one may develop any number of beautiful statistics that generally describe a player’s or team’s performance. But even in a game as simple as baseball (and I’m not knocking baseball), you can’t always make reliable predictions based on those measurements.

How many working parts are there in football? Run Ball (infinite ways), Pass Ball (infinite ways), Pitch Ball (CPJ’s way), Block High, Block Low, Pull Block, Motion, Different Formations, Reads, Audibles, Mismatches, Broken plays, Play Fakes, Penalties etc... And that’s just on offense.

I am a helluva engineer, with an impressive mathematics resume; though it should be noted that in the “use it or lose it” domain, I am much closer to “lose it”. One of my biggest pet peeves in the world is the way people use statistics for the purposes of confirmation bias. I can think of three specific ways off the top of my head that statistics can be misleading. These are as follows:

1. Cherry Picking or Falsifying Data – Everybody has seen this, its pretty self explanatory. An example of this is how the anti-vaccination movement makes its bogus correlations between autism and vaccines.

2. Reporting Bias – The very way statistics are reported can show an incomplete picture. An example of this is the CDC study that shows if you are exposed to 2nd hand smoke, you are 25% more likely to develop lung cancer. However, another interpretation is that the fraction of the population (ahem…sample) that develops lung cancer goes from something like 10 out of 100,000 to 12.5 out of 100,000. Is that really statistically significant, or are we just justifying our indignation with smokers?

3. Incomplete Analysis – Another way that statistics can be bastardized is by lumping a bunch of numbers together and drawing unproven conclusions with certainty. My example here is that people who have guns in their house are supposedly…I don’t know…say… fifty-eleven times more likely to be involved in an accident than a self-defense situation. Does that account variables like age, education level, NRA membership, safety training, regular practice, mental health history, etc..? No. So that statistic is meaningless to an individual.

This third way of misreporting statistical results is the most insidious I believe, because the first two statistical fallacies are easily dismissible. Incomplete Analysis on the other hand doesn’t mean simplistic analysis. Numerology is quite sophisticated, after all.

Jabs aside, Mr. Mathlete certainly has developed some interesting and presumably accurate charts. For those of you who would like a translation of his work, my addition to his basic chart should help you.

The Mathlete’s Statistical Conclusion

From Mathlete

So, go for it every time on 4th and 5 if there is less than 55 yards to the endzone (except if, and only if, you are on the 24, 22, 17, 15, and 14 yard lines, in which case you kick it). Really? How many of you really believe that?

I’m picking on the Mathlete here a little, and I would be remiss to omit the fact that he says the areas around the edges are not precise. That said, ~100% doesn’t need to be too precise. And don’t get cute people, we of course are leaving out end-game scenarios and freak situations (injuries on a preceding play for example).

But this analysis doesn’t include every circumstance, or really any specific situation. Oh, I almost forgot. He anticipated three objections and offers explanations in his post for the following:

  1. Doesn’t account for quick change momentum

  2. Assumes all defenses (or offenses I suppose) are average

  3. Doesn’t account for game specific situations

The Mathlete then shows charts for expected points per drive compared to actual points per drive to address item #1. His variables are for drives started by turnovers, 4th down stops, time outs, and normal drives. For item #2 he shows a variation of the chart for the Florida offense and Ohio State defense.

What does this prove? In the most convoluted and confusing way possible, he demonstrates that the strongest indicator of obtaining points on any given drive is field position, not how the ball was obtained. We’re all in agreement here. Maybe that’s why there are situations more than 0% of the time when you might consider punting from the 50.

So what else should be considered to further figure out when to go for it? When would you want to punt in mid-field and mid-distance situations? You’ve got to consider risk vs. reward. I love how people in favor of eliminating the punter position always say that points are equally important at all times in a game (except in end-game scenarios). I’m not so sure. Points are more critical when you are behind. Yes, you want to score every possession. But that isn’t the end goal. The MOST critical thing in the game of football is that you score more points than the other team before the game ends. If you are up 35-0 at half time, scoring points is still very important. If you are down 14-10 at the half, points are everything. If you are down 70-0 at any time in a game, working on fundamentals becomes the most important thing. Points be damned.

Am I saying you should punt if you are up big? No. I’m saying that it is one possible factor that may be considered within the scope of the situation. If its 50-50, punting doesn’t seem like a good idea. If it’s 3-3 late in the third, I might be punting from my own 45.

Say you’ve got a weird score at the half. You’re down 18 to 20. Then in the 3rd your offense just isn’t clicking. The defense made adjustments. The things you are trying to do aren’t working out, and you need to talk to your offense. Perhaps your QB isn’t seeing something and he needs to be coached on something. No headsets for players in college. Better live to fight another drive. Kick the field goal and get the lead. It’s not the 4th quarter but, economically, the opportunity cost of letting those 4 points go isn’t as hard to swallow.

Also consider that you might not want to show everything in your hand in the first quarter. This isn’t as big of an issue with someone like CPJ, because he’s coming at you with the same stuff regardless of down or distance. That kinda frees him up to do whatever he wants. But the vast majority of teams have specific plays designed for critical situations and for specific opponents. And points later in the game are typically more meaningful than points early in the game.

If the Mathlete is saying that more teams should run spread option so that they don’t have to punt, then I agree. Otherwise, maybe you consider punting more than zero.

What is the confidence you have in making the 1st down? Most offenses are not Florida. Most Defenses are not Ohio State. How do you make that distinction? See Mathlete’s graph again with my possible estimate of how certain you might want to be to make the decision to go for it. Note this is just a guess based on how precise the circles drawn in MS Word can be, so don’t hold me to an extrapolation.

Example Confidence Intervals of 90%, 60%, and 10%

From Mathlete

Based on an infinite number of factors, the sizes of the 90%, 60%, and 10% confidence intervals will vary wildly. And when you weigh that confidence (risk) against the expected points (reward), you intuitively know whether or not to go for it. Even if the averages are with you, the specific situation may not be.

But this gets us beside the point. No, the Mathlete has not provided us with standard deviations, we don’t know z-factors, we have no weighting, and I’m not even sure where the data comes from. But this is a football blog, not a math class. We need to realize that statistics can only take you so far. Concepts of risk / reward, and game theory are every bit as critical, and really more so. This game is not played on a spreadsheet (or in the blogosphere for that matter).

And hey, why does the Mathlete tell us to kick a field goal on long yardage inside the 30 yard line 100% of the time? I think that is just as unreasonable as going for it each time? Beware anybody who gives you Always and Never as your only options. We’re admittedly being picky here as the Mathlete has authored a very fine post.

In the ACC Championship game last year, I’m pretty sure that either team kicking a field goal on 4th and 8 from the 12 would’ve signed their death warrant. Speaking of Clemson, was it a wrong decision to fake the field goal in the first Clemson game from even greater distance? Even I liked that one. Likewise, consider what the Patriots did against Indy last year. Either way you slice it, you can’t make decisions in football off of list of averages.

If you start to include lots of qualitative variables, you might start to see a number of situations where it makes sense to punt, even when coach-fear is eliminated.

So in summary, coaches should go for it on 4th down more often. But they should use their head, sound strategy, and cool demeanor - not canned data. I don’t need any charts or graphs to tell me that. Even Hash is on board with that sentiment.....

1 comment:

  1. I think a very good analogy to this is the calculation of pot odds (and implied pot odds) in poker. You are calculating the percentage of the current pot that you can expect to win against the amount of the bet that you are going to call (or raise). If you are required to put in less chips than you can expect to win with your hand accounting for the texture of the flop, you should call or raise.

    I believe that this is particularly apt in that in both poker and football in that you as the coach or the player are expected to calculate odds with incomplete information. In poker, you can estimate what your opponent has based on the betting pattern; in football, you can guess what kind of defense that you will be running against based on the distance required, the texture of the game, and the tendencies of the coaches.

    Poker and football decision-making have one other thing in common; the situation that you are in can wildly swing your strategy. When playing in a poker tournament, when you are near the bubble where the payouts begin, it would be unwise to allow all of your chips to go into the pot unless you feel that you have much better odds of winning than at any other time in the tournament. Similarly, when the number of possessions in the game is diminishing and there are points to be gained by kicking a field goal, or giving your defense a better chance at a stop when you are up by one score or less, you would need more certainty to go for it on fourth down.

    One huge difference between football and poker, though, is the amount of opportunities that you can expect to receive in a given game. In poker, you can play hundreds of hands in the time it takes to play a football game. This allows the odds to "even out." Football, however, only gives your offense 10-15 opportunities (more or less). You become much more susceptible to random outcomes and luck when you may only face the situation a handful of times in a game. Over the course of a season, all of these factors will even out. But in a single game, that decision, and the "luck" involved, can change the outcome without a number of subsequent opportunities to balance out the odds.

    Particularly in college football, where a single loss can be the difference between a championship run and a mid-tier bowl, it doesn't surprise me that the coach whose livelihood is at stake would take the most conservative route. Put it on the defense, "let the players play," and if you lose that fourth down call doesn't put you on the hot seat. This is very similar to the NFL, and possibly more so.

    One final thing that Mathlete points out, and it is something that I have seen with CPJ's offensive play calling, is that if you have a mindset that you will be going for it on fourth based on where you line up on first down, you have opened up your play calling significantly. The pro-style offense, with a punt first mentality, is run or play action on first, run on second if you picked up four or more, pass if less, and basically pass for the first on third if you have more than a couple of yards. This allows the defense to pin back its ears and makes it harder to get a first down. If you begin on first knowing that you have four plays to make a first down you can step out of this box and really disrupt the defensive rythym, opening the door for "big" plays. Where I notice this most with CPJ is when it is third and 7 or 8 in plus territory. He can call his base triple option play, which is the most practiced and best performing with the offense. A successful play gains 4-5 yards, making the same play perfect for picking up the first. This allows the offense to run its best plays in these critical situations instead of depending on a pass which has not been nearly as successful on third and fourth downs.