What is an Outlier?

An outlier is a data point that sits far from the rest: much higher or lower than the pattern the other values make. It matters because one extreme value can drag an average a long way from where most of the data sits. A decision built on that average is really built on the outlier. The number alone does not say what to do with it. It may be a mistake, a one-off, or the start of a new pattern, and only checking the story behind it tells you which.

An outlier is a value that lies unusually far from the others in a set of data. Its distance is what defines it, often set as more than two or three standard deviations from the middle. What to do about it is a separate question, and the answer comes in three kinds. The outlier may be an error: a decimal in the wrong place, a broken sensor, a figure typed twice. Errors get corrected or removed. It may be genuine but caused by something that will not repeat: a one-time event that inflated one reading. Genuine one-offs get kept in the record and left out of any forecast, because they say nothing about the ordinary run of things. Or it may be genuine and the first sign of a change, in which case it is the most important value in the set. Nothing in the number itself distinguishes these. Two datasets can hold the same outlier for opposite reasons, so the only honest response is to go and find out what happened. What makes this urgent is that averages are sensitive to extremes. A single very large value pulls the mean towards it, which is why a room with one billionaire has a meaningless average income and why medians exist.

A factory sells around 100 units a month for eleven months, then 900 in one month. The average is now 167, a level no actual month resembled, and someone asks whether to expand the plant. Before answering, find out what the 900 was. If one customer's order was keyed as 900 instead of 90, fix the number and forget it. If the order was real but a competitor's plant was shut that month and their customers bought from you once, keep the record and leave it out of the plan. No expansion. But if that customer has moved all their business to you for good, the 900 is the new normal and the plant does need to grow. Same outlier, three different actions, and the number alone could not choose between them.