Outliers

One of my favorite stats of all time is that Mariano Rivera had a career BABIP of .265 (league average over that span was .298) and a career HR/FB rate of 6.2 percent (league average was around 10.) The BABIP mark was the lowest of any pitcher over that 20-year span (minimum 1000 IP), and the HR/FB was second-lowest. How could anyone get so lucky as to be No. 1 and 2 in two different, “luck-based” metrics?

Rivera is an outlier. Given the movement on his pitches (really one pitch) he was able to induce weak contact so consistently the balls in play against him were easier for his fielders to handle, and the ones in the air didn’t typically travel far. He was hard to square up, and he broke a lot of bats too. So not only was Rivera a statistical anomaly, but we also can understand the mechanism by which he did it.

Inducing weak contact is not a skill most pitchers have. Even great pitchers often rely on missing bats and not issuing free passes rather than limiting what happens when batters make contact. For this reason, it is indeed the case *on average* that BABIP and HR/FB rate are largely luck. But what’s true on average is not true for outliers — otherwise they would not be outliers.

It’s tempting to believe outliers should be ignored — after all, they are the exceptions to the rule, and “hard cases make bad law.” But outliers disproportionately drive outcomes as the rule is usually priced in, and anomalies are not. Sixth-round draft picks don’t win seven Super Bowls, 6-9, 250-pound small forwards don’t average seven assists per game over 20 years, kickers don’t make 90-percent of field goal attempts over 12 years, despite attempting many of them from well beyond 50 yards.

Outliers break models, they ruin the smooth distribution curves that make us feel like we understand what there is to know, they don’t dutifully regress to the mean over time the way to which they’re supposed. But we watch to find the outlier, not the average. We want to witness greatness because it reminds us of what’s possible rather than what’s likely.

In short, so much of what passes for “sharp” analysis has it backwards — the outlier is the signal, and the average, upon which we calibrate so much of our understanding, merely noise.