The seismic station in the hills records ground movement every day of the year, and for three hundred and sixty-four days the record is almost dull. The readings sit between one and three on the station's scale, day after day, and the most common reading by far is two.
Then, on one afternoon in March, the needle moved to sixty-one.
Look at what that single day does to the year's statistics. The most common reading is still two; it was two before March and it was two after. The middle reading, if you line the whole year up in order, is still two. Neither of those numbers notices the earthquake at all.
The range, though, went from two to sixty. And the year's average, which had been sitting quietly at about two, was dragged upward by a single value some thirty times the typical reading.
A student on placement at the station asked whether the March value should be excluded as an outlier, since it was making the year look unrepresentative.
The station's director gave the answer he gives every year to every student. The other three hundred and sixty-four readings are the ones you could have guessed. This station was built, staffed and paid for because of days like that one. "An outlier is not noise by definition," he said. "Sometimes it is the whole point."