As I begin typing these words, the NFL season has 25 hours and one game to go. I have listened to about 13 days of chatter about who will win that one game and why. And in general, a season’s worth of prognosticating is fresh in mind.
Brian Burke argued, in a NYT blog post I discussed last week (below), that the best experts typically get a third or more of their picks wrong, even simply attempting to pick straight-up winners where many teams each week will be favored by several points. Why?
I suspect that experts and regular fans alike make bad picks for somewhat different reasons in each major sport.
First, here are some reasons why we are often likely to get picks wrong in most sports, and in other areas as well — such as picking stocks, predicting election outcomes a year in advance, and so on.
(1) Hindsight bias. Following Burke’s suggestion, below, we are inclined “to think that past events were more predictable than they really were. People also tend to whitewash their own predictions, recalling them as more accurate than they truly were.” Because of this, we are less like to adapt the “theories” and data we are using to make our predictions.
(2) Recency effect. “The tendency to weigh recent events more than earlier events.” Sometimes the fact that a team did well in its last game is an indication of how well it will do in the next game. But sometimes there were very particular reasons for that success, and we should not ignore how poorly they did in the previous several games and what this tells us about their future chances.
(3) Halo effect. In the case of sports predictions, this might be thought of as the tendency to over-rate glamor teams (especially the ones in big media markets) or teams with outsized stars (even if they are not statistically superior to some rivals).
(4) Failure to recognize the role of randomness in previous results. While watching a game we appreciate how crucial one catch, drop, penalty, or missed kick was in determining the outcome. But within a day or a week we treat that result as if it were the result of solid, stable underlying tendencies. We then overestimate how “powerful” these tendencies — e.g. a quarterback’s or coach’s genius — will be in determining future results.
These four types of cognitive bias affect how we predict many things in life. But at this moment it is worth thinking about some particular reasons NFL punditry is so difficult.
(5) Following on (4), above, there seems to be a massive amount of randomness going on in football, with 22 players and a ball in motion on every play. Is it possible there is an order of magnitude more probabilistic variables at play in football than in, say, baseball? I’m not sure how you would count. At the same time, it is much harder to capture relevant statistical tendencies in football than it is in, say, baseball. With less of the game susceptible to number-crunching, it is more open to our cognitive biases and our natural attempts to read an illusory order into the randomness.
(6) Relatedly, success in football, on either the offensive or defensive side of the ball, requires more coordination among teammates than in other sports (and not just because there are more players on the field of play than in other sports). A soccer team can hold its own, and even pull ahead, when one of its players have been red-carded. But the best NFL team could never beat the worst team if it had to play 10 against 11 for even half or a quarter of the game. Similarly, any given play is likely to break down if just one of the 11 players blows his specific assignment. This will happen less often with better teams (or we call them better because it happens less often), but one single breakdown by one player on one play can have game-changing consequences. I suppose that statement is true in all sports. But in football that crucial breakdown may take place far from the action, so to speak. (One defender’s stance before the snap may give away information to the quarterback that allows him to complete a touchdown pass that beats another defender on the other side of the field 10 seconds later.) In short, it seems truer in football that the chain is only as strong as its weakest link. How can anyone predict which team will have more of broken links in a given game?
(7) Forgetting that “it’s all about matchups!” (This observation comes from my wife — who, by the way, knows much, much, much more about statistical inference and strategic rationality than I every will.) The most interesting pundits will often focus on specific “matchups” in an upcoming game. And so will coaches preparing their teams for the game. A matchup might involve a particular offensive lineman against a defensive lineman, or a particular wide receiver against some cornerback, the quarterback against the middle linebacker adjusting defensive formations, etc. In advance of the game, the experts focus on how much rides on some identifiable matchups between specific players with specific traits that may or may not be successful against their particular adversary. And yet after the result of the game is in, we treat the winning team’s success as if it were just about sheer superiority. And expect it to do equally well against the next opponent; despite the fact that the specific matchups may be very much less favorable in that case.
(8) There is incredible parity of talent across NFL teams. Few other sports have made it more difficult for teams to have significantly inferior players and coaches to work with. Consider some equalizing factors: all top incoming players are drafted in sequential order with the worst teams drafting before the better teams; moreover, the teams basically pool much of their information about incoming players (e.g. testing them in collective “combines”); there is a salary cap that most teams meet, which means that all teams have about the same amount of money to spend on talent (although, of course, they can draft, trade, and attract free agents unwisely, or use too much of their capped salary pool on the wrong players). I think this means that if even the best team is not mentally, physically, and emotionally prepared for a game against a very bad team (well, at least one not as bad as the Lions), they are much more likely than we would ever imagine to lose. Certainly much more likely than a leading top-division soccer team in most European leagues against a team facing relegation in its same division.
(9) There is also a powerful mechanism for parity in coaching talent and a very short-lived advantage from any valuable “intellectual property” a coaching staff may have developed. Subordinate coaches (e.g. offensive or defensive coordinators, and the coaches below them) on successful teams are quickly snapped up, and often promoted, by other teams. So any secrets of success are quickly dispersed through the league, and no team can hold on to good coaching talent for long.
(10) And finally, prediction — especially for single games — is rendered more difficult in football by the greater degree of strategic rationality at play than in other sports. That is, by the amount of cat-and-mousemanship. In order to be successful you have to take into account how the other team is trying to beat you. And they will take into account how you are trying to take into account how they will be trying to beat you. And so on. This goes on in every serious sport, but I dare say to a much greater and more scripted degree in football. In other sports, a winning strategy may continue to succeed game-in, game-out, because other teams just can’t counter it. But in football, there are enough options on the other side of the ball that the opponent has the option of neutralizing your successful strategy at the risk of making some very different strategy more attractive for you. (E.g. stopping your successful running game by putting the safety into the box, thereby making it easier for you to be successful in the passing game.) Surely it is more difficult to predict who will win when teams are forced to try to win using very different strategies from one week to the next.
So, as for the Super Bowl tomorrow, who ya GOT?!