If you’re a football fan, and you’ve never read Gregg Easterbrook’s Tuesday Morning Quarterback columns at espn.com (or previously, nfl.com), you’re in for a treat. I got hooked on them a few years ago – smart, funny, insightful. Wonderfully entertaining.
But as you might guess from the title of this post, I come to bury TMQ, not praise him. These days his columns annoy me like all get-out. Part of this is that the act is wearing thin – the jokes don’t change; the lines are the same, just some of the names and teams have changed. But more seriously, I object to Easterbrook’s cavalier and probably dishonest use of statistics.
Case in point. Easterbrook has been on a crusade for some number of years to warn the world about the dangers of the blitz. His latest diatribe can be read at
http://sports.espn.go.com/espn/page2/story?page=easterbrook/060912. Sometimes it’s hard to tell what exactly his point is, but my understanding is that he believes that calling a blitz is inferior to using a standard defense, and therefore that teams shouldn’t blitz; at the very least that they should blitz much less than they do. Certainly, TMQ believes that if a team blitzes, and gets burned by it, then it is worthy of ridicule. This theory wouldn’t bother me much, except that Easterbrook keeps using his column as a bully pulpit for it, with a very smug, condescending tone and using statistics in wholly inappropriate ways. Grrrrrr.
Before I get to the statistical arguments it would be good to note that you don’t even need to hear them to tell what the actual state of affairs is. Blitzes, while not infrequent, are called much less than half the time. So obviously, defensive coordinators throughout the league believe that the blitz is less effective on average than not blitzing. At the same time, all defensive coordinators in the NFL do call blitzes, a fair percentage of the time so they must feel that the blitz has some worth. Whose judgment sounds better, that of experienced, knowledgeable people whose jobs are on the line, or some number-crunching armchair quarterback talking head?
But on to Easterbrook’s evidence. Every column has a section called “Stop Me Before I Blitz Again”, which has three or four examples from the previous week’s action where blitzes were ineffectual. Wow. So he cherry-picks a few plays and uses them to prove his theory. That’s as meaningless as it is predictable. Hey, why not run a “Stop Me Before I Pass Again” section, where you replay a few interceptions each week to prove that passing is bad.
He crunched some numbers one weekend, and determined that “On 158 long-yardage downs … offenses averaged 4.3 yards per play against conventional defense and 8.7 yards per play against the blitz and scored touchdowns on 2.6 percent of plays against conventional defense and on 6.7 percent of plays against the blitz. Offenses scored three touchdowns against conventional defenses, a 2.6 percent touchdown rate, and three touchdowns against the blitz, a 6.7 percent touchdown rate … This is pretty much a slam-dunk for conventional defense over the blitz.” (
http://www.superbowl.com/news/story/9156476)
Interesting statistics. But honestly, you don’t even need to think about them to know that there’s something rotten with his conclusion. Let’s take a wholly analogous situation and see where that kind of logic leads us.
The average pass play in the NFL yields what, something like seven yards (or thereabouts; I’m too lazy to look it up). And the average rushing play gains on the average of three-and-a-half yards. A higher percentage of passes result in first downs than runs; ditto for touchdowns. So, using Easterbrook’s logic, rushing sucks … slam-dunk. Teams, stop rushing! (By the way, TMQ actually applied a version of this logic in a basketball column. He noted that in an NCAA game last March, one team had a better average points-per-shot for two-point attempts than three-point attempts, and concluded that they should only have taken two-point shots the entire game.)
So the stats, accurate or not, are meaningless in the big picture. But while I’m here, I’ll take the trouble to point out the issues with them.
- These numbers are too coarse to make apples-to-apples comparisons. For example, the touchdown numbers (which are too small to be significant) are meaningless without knowing field position. More to the point, the statistics need to be broken down by the comparative strength of teams to be meaningful. In at least some cases, blitzes are called by weaker teams because their standard defenses are even worse.
- More importantly, an essential part of football strategy is to keep the opponent guessing. If you never blitzed, your opponent can take advantage of the fact. Even if the blitz is a suboptimal tactic (and maybe it is), you still need to call it relatively often to keep the foe off-balance. This, of course, is one fallacy committed by my run-pass example above.
- And all that is just a way of saying that blitzing has ramifications beyond the play at hand, which the statistics don’t take into effect. Another example of this is that some teams blitz because it means they get to hit the quarterback more often. If it’s a sack, great, but it’s still beneficial just to hit him, which can rattle him, physically or emotionally, and throw him off his rhythm on future plays.
When I was ten, I figured that with a 3.5 yard average rush, all a team had to do to get a first down every time was to run for three straight plays. Three times 3.5 is 10.5 yards, first down. Heck, if you’re willing to go for it on fourth down, you only need to average 2.5 yards. Surely a team can do that, right? Took me a while to discover the flaw in my logic, but then I was just a kid.
One more example of his use of statistics on the blitz and then I’ll move on. In the latest column, he says “If blitzing works better during the regular season than the playoffs, we'd expect to see the best teams blitz steadily less come January. And that is exactly what happened last season. During the regular season, the Steelers … on 32 percent of its 2005 regular-season defensive snaps, versus the league average of about 15 percent blitzing. As the playoffs progressed, Pittsburgh backed away from the blitz, dropping into conventional defense. Against Cincinnati in the first round of the playoffs, the Steelers blitzed on 18 percent of Bengals plays; in the second round, Pittsburgh blitzed on 25 percent of Colts plays; in the AFC championship, Pittsburgh blitzed on 16 percent of Broncos plays; in the Super Bowl, the Steelers blitzed on 13 percent of Seahawks plays.”
Now the theory that blitzing is less effective against good teams might well be true; it’s certainly plausible. I just don’t like the evidence marshalled here. Let’s see, the champion team, the best of the best, blitzed 32% of the time, more than most teams, and this is used as evidence by a guy who wants to convince us that the blitz is bad. Sweet! That’s pretty funny in itself. But consider that trend of 18%, 25%, 16%, 13%. I would hardly call that “blitzing steadily less”; in fact, consider the raw numbers. I’m too lazy to look them up, but I think that in a typical NFL game, each offense runs about 50 plays. This would put the number of blitzes at 9, 12-13, 8, 6-7. Wow, so Pittsburgh blitzes once or twice less often in a couple of games, and it’s a trend. Needless to say, there’s not a whole lot of statistical significance here in the technical sense. More likely explanations for the variation would come from game circumstances, or the fact that the Pittsburgh coaches decided that it would be more effective against certain teams than others.
Let me get off Easterbrook’s back and ask if his statistical case (i.e. by comparing yards per play) can be salvaged. A better approach would use that branch of mathematics called game theory, although I have serious doubts about its success. (
Update: I fleshed this out in a subsequent post. See
http://northern-flicker.blogspot.com/2006/09/football-and-game-theory.html.)
Easterbrook’s latest crusade is on something he calls the “Maroon Zone”, which is that part of the field that is not quite the red zone. Too close to the end zone to punt, too far to kick a field goal. His theory is that teams should go for it on fourth down more than they do – which I actually agree with. On a related note, he also likes to taunt losing teams for not going for it enough late in the game, when they have nothing to lose. Again, I think he has something of a point. But of course, this instantly raises the question of where you draw the line. Fourth and seven on the opponent’s 36-yard-line? Go for it, or punt? Fourth and two on your own 35, five minutes left in the game, you’re down 35-21. Go for it, or punt? Fourth and six on the opponent’s 26, ten minutes left, you’re down 24-10. Go for it, or field goal?
Easterbrook answers that question in his usual facile, cherry-picking way. If the team punted, and it turned out that the opposing team scored on their ensuing possession, then he points out that the punt was a bad choice. If they went for it and made it, he congratulates them on following his advice. Counterexamples, such as punting, bottling up the opponent’s offense, and getting the ball back with good field position are ignored. Well, that’s ummm, hard to argue with, but then that sort of logic doesn’t really lend itself to making good decisions in the future.
Interestingly enough, these sorts of questions do lend themselves to mathematical analysis, although such analysis wouldn’t be easy. Easterbrook has never even mentioned that there is risks to his strategy. That risk is blindingly obvious, but I’ll mention it for the sake of completeness. If you go for it and fail, you give the opponent better field position. Even when you are down at the end of the game, you always have a shot (however small) of winning by playing it safe, but that shot gets worse if you give your opponent get easy chances to score.
Anyways, you could build a model for analyzing the tradeoffs with assumptions like:
- chance of making a first down on 4th and n yards to go
- chance of scoring once you make the first down
- probability distribution of opponent’s offense given that they start on the x yard line
- probability distribution of your offense given that they start on the x yard line
- probability distribution of time of possessions, for both teams
(By probability distribution, I mean a spread of outcomes. A sample distribution could be that, for the case that your team starts on the 10 yard line, then there’s a 10% chance of an eventual touchdown on that drive, 15% chance of field goal, whatever percent chance of punting/turnover and the opponent getting the ball on the m yard line, etc. You really need to use entire distributions here and not simply go with averages to get valid results.) I would be hesitant about applying the results of such a model in any detailed sense, but it could give you some rules of thumb on when it’s better to go for it. (By the way, I kind of did something like this a long time ago, in a very abstract way; maybe I’ll write a post about that someday.)
While I’m at it, let me dispute another of Easterbrook’s notions, that all that matters is winning or losing, and the score is irrelevant. So, for example, say the situation is 4th and three at the opponent’s 20 yard line, and you are down 28-0, early third quarter. This is your best drive of the game. Easterbrook would argue that you have to go for it here, instead of kicking the field goal, as that will give you the best shot of winning the game. Other examples that he’s given include going for it in your own territory early in the fourth quarter.
I disagree for several reasons. One is that, taking my example, kicking the field goal builds momentum, helping your team down the stretch, whereas coming away with nothing deflates it. This is the reason many coaches give in explaining their conservative choices.
My second reason is a bit long-winded. Consider this example. You are down 21-0, and have the choice between strategies C and R. Strategy C is conservative, but unlikely to win the game. Using this strategy, you have a 5% chance of winning 24-21, and a 95% chance of losing 21-14. Strategy R is riskier, but has a higher risk of winning. Here, you have an 8% chance of winning 24-21, a 10% chance of losing 21-17, and an 82% chance of losing 50-0 (since you go for it on fourth down often). Sure, winning matters, a lot, but losing big has its own repercussions. It can hurt the psyche of your team, detrimental to their performance next week. If you’re the coach, it’s something your bosses could feel humiliated about – not a good way to keep job security, especially if you take these risks week after week.
In fact, this scenario played out once (sort of), with most interesting results. November 23, 2001, Colorado versus undefeated Nebraska. Colorado rushed out to an early 35-3 lead in the second quarter. Nebraska came back to make it 42-30 at the start of the fourth quarter. Then the pressure of being behind late caused Nebraska to abandon their bread-and-butter rushing game, and throw the ball recklessly. Turnovers (and hilarity) ensued, and Nebraska got blown out, 62-36, their worst loss in decades. If they had played it close to the vest, who knows what would have happened – they probably still would have lost, but by a respectable amount.
The blowout fallout was significant. Nebraska was, if not a shattered team, a team whose confidence had been severely undermined. They still maintained their #2 BCS rating (causing a furor in itself), but were just marking time until they could get blown out again in the championship game with Miami. In fact, Nebraska’s self-image has never returned to where it was before that game, and that score was probably a significant factor in the firing of their coach two years later. Meanwhile CU surged to a Big-12 championship victory over big-game-choking-Texas. They spent the next few weeks arguing that they were the real #2 team in the country, despite two losses on their record. Their protestations continued right up until their bowl game, when they were summarily dispatched by an Oregon team that was altogether stronger in every way.
Ok, I guess that got off-topic. I do like to reminisce about the good old days. The point (whether that story illustrates it or not) is that margin of victory is not irrelevant. It can better to marginally reduce your odds of winning a game that you’re probably not going to win anyways to keep the score close.
Back to Easterbrook. I think I see what he’s doing. As a professional “thinker”, you like to come up with ideas that nobody else has thought of, or at least has published, and make them yours. I can see him dreaming of some future History of Football Strategy textbook passage:
… The evolution of football strategy had come to a standstill by the year 2000, until it was revolutionized by the brilliant amateur Gregg Easterbrook, who decisively proved that the blitz had been massively overused by the entire professional and collegiate ranks. Moreover, Easterbrook’s Maroon Zone Theory gave teams an insurmountable tactical edge, at least until its adoption became standard …
If you think I’m exaggerating, note that one of TMQ’s columns is titled “New Key to Success: Maroon”. Yeah, right. At best, if you did the math, which Easterbrook doesn’t, and knew the right way to play situations, you could probably help yourself to maybe an extra point a game on average. A better key to success might be improving your running, passing, or defensive skills.
Back to the main point, it doesn't bother me that he's doing this. But when you propose a theory, you ought to weigh the evidence impartially to determine whether your theory is true, not indulge yourself in post-hoc reasoning, mine the statistics to find some that support your theory, and ignore the rest.
Sorry for such a long post. I usually don’t bother wasting my time pointing out why people’s math is crap, but this dude got under my skin.