mean vs median vs mode; one outlier can drag the mean further from anyone's actual experience
The average does not lie about arithmetic; it lies about what is typical. When a single value is far above or far below the rest, the mean gets dragged away from the center, so the number it reports may match no one in the group at all.
The arithmetic mean is not designed to find the most common or most central value. It is designed to answer one question: if every member of the group were given an equal share of the combined total, how much would each person get? That is why a small number of enormous values can dominate. The mean must distribute every dollar, every second, or every defect across everyone, including the extreme cases.
A quick hypothetical: five salaries are 30,000, 35,000, 40,000, 45,000, and 500,000. The total is 650,000, so the mean is 130,000. Yet no one earns 130,000, and four of the five earn far less. The median, in contrast, sorts the values and picks the middle one: 40,000. That matches the experience of the third person in line, who is the only one in the exact center, but it is also far closer to what four of the five actually take home.
Most people hear "average salary" or "average wait time" and mentally translate it as "what a normal person gets." That translation fails whenever the data is skewed. Skewness means the tail stretches much farther on one side than the other. Income is left-skewed in the sense that a few values are extremely high. House prices are similar: a handful of luxury sales can lift the mean above what a typical buyer sees. Web page load times and insurance claims behave the same way, with rare but massive values inflating the mean.
The classic error is to assume the mean is simply the midpoint of the data. It is not. The midpoint, properly called the midrange, takes the smallest and largest values and halves their sum. In the salary example above, the midpoint would be 265,000, which is even more misleading. The mean at least weighs every data point. But it still obeys the extremes instead of resisting them.
If the data is roughly symmetric, the mean and median sit close together, and the mean gains an advantage: it uses every value efficiently. Heights of adult humans, standardized test scores, and repeated measurement errors tend to form a bell-shaped curve. In those cases, the mean is stable and describes the group well. It also matters when the total itself is the point, not the individual. If you want to budget for total payroll, the mean times headcount is exactly what you need, regardless of how unevenly the salaries are spread.
Do not use the mean alone for anything with a heavy tail or a hard floor at zero. It will mislead. The antidote is not to abandon all single numbers but to choose the one that answers your question. If your question is "what is the total divided by the count," use the mean. If your question is "what do people actually experience," use the median. If your question is "which value occurs most often," use the mode. The average is not evil; it is simply a tool with a specific job, and it fails when applied to skewed data without a second thought.
Cram The average tells you what is normal, right?
Rep No. One extreme value can drag the average far from what anyone actually experiences.
Cram How does one number change everything?
Rep Ten people in a room. Nine earn 40,000 a year. One earns 40 million. The average is over 4 million.
Cram But nobody in the room makes anywhere near 4 million.
Rep Exactly. The average is correct but describes nobody. That is why the median exists.
Cram What does the median do differently?
Rep It lines up every value and picks the middle one. The billionaire is at one end, the middle person is still around 40,000.
Cram So the median resists the extreme values.
Rep Yes. It tells you what a typical person actually experiences. The average just divides total by count.
Cram When does the average actually make sense?
Rep When the data is symmetric. Heights, IQ scores, measurement errors. No extreme outliers.
Cram So whenever I see average, I should ask what the median shows.
Rep Right. And check the mode too. The most common value tells you what is actually happening.