Here's the Slate link. Here's the Deadspin link. This post is a little more Slate's speed than Deadspin material.
The first fallacy is called base rate neglect. This is an error all people tend to make, even experts. The most common example is when doctors are asked to estimate the chances of someone having a disease given certain facts. They almost universally ignore the base rate of the disease—that, say, only one in 100,000 people in the population will ever contract it...
Similarly, when coaches are asked in postgame press conferences about fourth-down decisions, they discuss only the particulars of the situation: how successful they were on the two short-yardage plays earlier in the game, the nagging injury to the left guard, and the success his defense has had so far that day. What they always ignore is the base rate of success. But it's the most important piece of information. It's the one thing you'd want to know as a head coach in such a situation...
Good article. This explains some of the poor 4th down decision-making, but I think there is also a more basic cause, which could be seen as a form of the "anchoring" bias or herd behavior. The default position on 4th down forever has been to punt, and that's what coaches have always seen and been taught to do, so they place a much higher value on punting than they should. I love Brian's thought experiment about what coaches would do on 4th down if the punt never existed and one day was introduced as a rule change.
I live in ATL and when Cam came in on 4th down I thought it was game over. When CAR tried to draw ATL offsides I knew they weren't going to run a play. They would punt and ATL would have a chance to win.
If you're a CAR fan you should be hoping that your team does what the opposition fears. In this case the Falcons were praying that Cam didn't run a QB sneak or worse yet: spread out the field and move the LBs away from the line of scrimmage.
I firmly believe that going for it was the right call, but I am skeptical of the numbers in that ESPN link. The implication is that the Panthers have an 83.5% WP if they go for it and 57.4% WP if they punt, yet a 90.9% WP overall. Huh? There is no way WP(total) can be higher than both WP(punt) AND WP(go for it) unless that delay-of-game dealt a massive blow to WP(punt).
Nice article, Brian. I would add a couple of other elements as probably playing a role:
1) the cost of failure when going for it are immediate, clear, and painful -- the opponent takes possession, often in good field position -- while the downside of punting is largely invisible (a would-have-been conversion that fans never see).
2) the coach's incentives are skewed here. He will receive 100% of the blame for a failed 4th down attempt, because it is seen as unconventional. But credit for successes will likely be shared -- with the players involved in a successful conversion, and players involved in any scoring play that may follow. A coach will get some acclaim for a gutsy 4th down call that works out, but not nearly the level attention that a failure will draw (in a sense, I suppose this pattern is a kind of loss aversion on the part of the fans/media).
The base rate of success is great and all but theoretically, what if a team like the Patriots who do go for it often have certain situations scripted so that the Patriots will ONLY go for it, if the presnap read is favorable otherwise they will take the delay of game or time out if they are unable to draw opponent offsides.. Then you are not looking at the actual stats of what an average team can expect on 4th down. The teams that go for it more often are also more likely to be more prepared, and have an advantage, otherwise they might bypass the opportunity. So when looking at the base success data you could potentially most likely looking at the best 50% of the teams in the best 50% of defensive looks. The conversion rate would no doubt be inflated.
Similarly, teams that practice surprise onside kick probably only do it because they observe a certain player or group of players cheating and leaving early and perhaps they even have a certain "tell" much like spotting a poker bluff has a much higher success rate given certain "tells" as such the onside kick data could be greatly inflated.
Finally, teams that are behind in the football game by 2 scores going for it may win thebattle, but lose the war. The teams on defense playing against them might be more concerned about giving up a long play and giving the team a significant chance at winning than they are of giving up a first down and keeping the drive alive. The defenses don't mind giving up more first downs, because it beats the alternative of even a small increase in the chance of giving up the big play. Intuitively things done by coaches may be right, even when they appear wrong.
Nevertheless, the base data is still extremely important, and certainly preparing a team to be more aggressive and go for it provided you get one of the more favorable defensive looks in those situations is advisable. However, the real question one should determine is WHAT percentage success rate does one NEED to have in order for the decision to be correct. That would be a more interesting study, more actionable data, and if the QB was area of it, he can try shifting guys around, using motions and audibles of formation until he gets a look favorable enough to justify going for it. If for example one only needs to convert 40% of the time for it to be worth it, just to be "conservative" the team could go for it if the QB estimates a 50% chance (as people tend to be overconfident in their own abilities) to provide a margin of safety. I don't really think that coaches are making a huge mistake in terms of estimating just how successful they will be, they most likely just view the consequences of not making it much greater than they really are, because they don't know the actual percentages needed to convert to justify going for it.
Ideal poker strategy may vary drastically if one correctly identifies his opponents strengths and his own strengths as well as weaknesses of both. Although folding 72 offsuit would for example seem to almost always be correct, if you identify a playerwho can be easily bluffed, and who is likely to go on tilt if you show a bluff, and you have a tell that he is weak that is reliable, even what seems like the bad play can be right.
I think there is a certain irony when coaches say things like "You gotta trust your defense can make the stop and punt" when in fact, even if you fail to convert, the defense has the opportunity to make the stop.
Jane Risen, Tom Gilovich, and Richard Thaler have a paper (which appears to still be unpublished) where they argue that this type of error in decision making is due to sudden death aversion. Their example, from basketball, is that teams which are down by 2 at the end of a game should take more three pointers, because a made two-pointer only extends the game into overtime and gives you a 50-50 chance of winning.
Their argument, basically, is that decision-makings want to avoid decisions which could immediately turn out very badly. They try to maximize the chance that they'll stay alive (or that they'll still be in at least pretty good shape).
Oh, definitely. I've long said that coaches of trailing teams aren't coaching to win. They're coaching to delay elimination as long as they can, hoping for a miracle along the way.
I think what is needed is a Mike Periria type analyst who is asked to comment on all 4th down decisions. Add this to a tally on ESPN of ep lost and Win prob lost by each 4th down decision per team during a game and over a season.
Of course it would become monotonous as on every 4th down the analyst would say they should have gone ofr it on 4th on 1. But that is the only way to change attitudes.
The article draws on the ideas in "thinking fast and slow" by kahneman. So we need to pitch the decision in terms of LOSS of points or win prob due to the coach's conservatism.
In terms of the argument that the average is less important than the specifics of the team and their performance in the game so far. I think this is the opposite of the lake wobegone effect as it means that every coach/fan/TV analyst thinks his offense is below average and the opponents defense is above average but this cannot be reality.
Brian, I love this article. I had never heard of Prospect Theory before but I think it explains the lion's share of the gap between optimal and actual risk/reward strategies in life.
However, I disagree with you when you write that fans aren't susceptible to the same weakness as coaches because it isn't the fans backs on the line. I think that fans are every last bit at susceptible to the same thing, including me.
I'm a huge Giants fan, and I think back to the Ahmad Bradshaw reluctant touchdown in SBXLVI. At the time, I agreed with the decision. I didn't agree with it because I thought it maximized the odds of winning, but because the thought of losing on a botched 18 yard field goal at the end would have been unfathomably gut wrenching. Seeing our defense get carved up by Tom Brady in the last 45 seconds would have been massively preferable, albeit far more likely.
I even realized at the time that my thinking was emotional and flawed, but I couldn't help it. How can I hold a coach to a higher standard of decision making than I myself am capable of?