What Anscombe’s Quartet Teaches About Data Analysis

You’ve probably seen averages, variances and correlation coefficients in maths class. They’re handy shortcuts to describe a bunch of numbers. But what if I told you four totally different datasets could share the same average, spread and correlation? That’s exactly what statistician Francis Anscombe cooked up in 1973 to prove a point: never trust summary stats alone. This post uses Anscombe’s famous quartet to dive into why good data analysis is as much about asking the right questions and drawing pictures as it is about crunching numbers.

Where did this come from?

In the early 1970s, Francis Anscombe—an English statistician—noticed his students relied too heavily on basic formulas and ignored the actual shape of the data. To drive home the point, he created four small datasets, each with identical means, variances and correlation between x and y values. When you graph them, though, they look wildly different—one is linear, another curves, a third has an outlier, and the last sits in a straight vertical line. Anscombe published his quartet in 1973, and it’s still a classic reminder to visualise before you theorise.

Where you’ll see this in real life

1. Business dashboards: A CEO might spot an upward sales trend in averages, but a hidden outlier order or seasonal dip could be masked. 2. Medical trials: Average recovery times can hide a subgroup that reacts badly to a treatment—visual checks can reveal that. 3. Sports analytics: A player’s average score might look steady, but plotting game-by-game can show hot streaks or slumps. 4. Election polling: Overall percentages often ignore regional swings; mapping poll results helps campaigners target key areas.

A common misconception

One trap every new data analyst falls into is thinking that numbers alone tell the whole story. You might calculate a perfect correlation and assume you’ve found a rock-solid relationship—until you see a scatterplot that shows a curve or a single outlier driving the result. Always plot your data first, ask what patterns or exceptions you notice, and then back it up with the right statistics.