Misleading Average

In plain terms

"Average" is doing a lot of quiet work in English. Usually it means the mean: add everything up, divide by how many things there are. But there's also the median, the value sitting in the middle when you sort them, and the mode, the value that comes up most often. On a symmetric spread of numbers the three land close together. On a lopsided one they can be wildly far apart, and only one of them is telling you about a typical case.

The mean is the one that gets quoted, and it's the one that breaks. It has no defense against extremes. A single enormous value drags it upward past almost every number in the set, so the "average" ends up describing a person or a case that doesn't exist.

Income, wealth, company size, page views, customer spend, sick days, wait times: these are all skewed distributions, meaning a long tail of large values on one side. In every one of them, most cases sit below the mean.

Why it matters

A summary statistic is a compression. You throw away the shape of the data and keep one number, which is fine as long as the number you kept answers the question being asked. The failure happens when the question is "what's a typical case?" and the answer given is a mean pulled off-center by the tail.

The mean isn't wrong, and it isn't a lie. That's what makes it effective. Anyone can check the arithmetic and it will hold up, which is why the number survives scrutiny that the impression it creates would not.

Canonical example

"Average salary at this company is $145,000. We pay extremely well."

Ten employees. Nine earn $50,000 each and the founder takes $1,000,000. Total payroll is $1,450,000, and divided by ten that's a mean of $145,000. The claim is arithmetically perfect.

It's also useless to anyone asking what they'd earn there. The median is $50,000. So is the mode. Nine of the ten employees make about a third of the "average", and no one on the payroll earns anything close to it. One number, one outlier, and a recruiting pitch that falls apart the moment a candidate sees a paystub.

Watch for the same shape in "average customer spend", "average time on site", and any statistic about a small group containing one very large member.

Counter-example (not a fallacy)

"Average revenue per customer is $145. We have 10,000 customers, so we should plan for roughly $1.45 million."

The mean is exactly the right tool here, and the distribution being skewed doesn't matter at all. Mean times count recovers the total, which is the question being asked. The median can't do that: multiplying a median by a headcount gives you a number with no meaning. Forecasting totals, budgeting, and anything involving a sum is mean territory.

The line: is the number answering a question about a typical case, or about a total? The mean is honest about totals and unreliable about typical.

How to fix it

If you've been linked here, give the reader a second number. A median alongside the mean shows the skew instantly, and the gap between them is itself informative. Better still, add a spread: the 25th and 75th percentiles, or the range, or just the largest value if one case is doing all the pulling. When the distribution is bimodal, two clusters with a gap in the middle, no single summary works at all, and the honest move is to show the shape or split the groups. "Half our staff earn under $50,000, and pay ranges from $50,000 to $1 million" is a sentence nobody can accuse you of spinning.

If you're on the receiving end, ask for the median and the range. "Mean or median?" is a short question that does a lot of work. If the two are far apart, the distribution has a tail, and the conversation should be about the shape rather than the single number.