For a right-skewed column like household income, why does the mean exceed the median?
answer
- one uses sizes, one uses positions
- the tail runs to the high side
- moving the top value changes nothing at rank
- mean gets dragged, median stays put
- mean above median flags right skew
basics
~20 sThe mean sums every value, so a long right tail of very high incomes pulls it upward. The median depends only on rank, so extreme values barely move it. Mean above median is the usual signature of right skew.
solid answer
~50 sHousehold income is bounded below by zero but unbounded above, so a small number of very large incomes stretch the distribution far to the right. The arithmetic mean adds every value, so each of those large incomes contributes its full magnitude and drags the average up. The median is whatever value sits at the middle rank position: moving the top earner from 500k to 5m changes no one's position in the ordering, so the median does not move at all. That asymmetry is why `mean > median` is the standard quick signature of right skew, and why an income headline should quote the median. The median answers what a typical household earns; the mean answers what each household would get if total income were shared out equally. Treat the rule as a dependable heuristic for unimodal data rather than a theorem, since contrived or multimodal shapes can reverse it.
go deeper
Be ready to state the direction from memory and say why: mean above median means a tail on the high side, because the mean feels each value's size while the median only counts positions.
Explain the mechanism with a small worked list, and show that changing the largest value moves the mean but leaves the median fixed. Know that mean equal to median suggests symmetry, not normality.
Demonstrate reporting judgment: median for typical experience, mean when the total must reconcile, both when describing shape. Expect to defend the choice to a stakeholder who wants the flattering number.
Own the standard. Decide which centre your organisation's headline metrics use, write it down, and make sure dashboards do not silently switch between mean and median across teams so numbers stay comparable.
## The three centres are three different questions A measure of centre compresses a whole column into one number, but the three classic measures compress it in different ways, and the difference is exactly what this question is about. - **Arithmetic mean**: add all n values, divide by n. Every observation contributes its *magnitude*. - **Median**: sort the values and take the middle one (for even n, the average of the two middle values). Every observation contributes only its *rank*. - **Mode**: the most frequently occurring value. Every observation contributes only to a *count*. Because the mean uses magnitudes and the median uses positions, they disagree systematically whenever the distribution is lopsided. ## Why income is lopsided A household cannot earn less than zero, so the distribution has a hard floor. There is no corresponding ceiling: incomes run up through six and seven figures with a thin but real density all the way out. The result is a distribution with most of its mass packed near the low end and a long thin tail extending to the right — a **right-skewed** (equivalently, positively skewed) shape. Now watch what each statistic does with that tail. Suppose ten households earn, in thousands: 20, 25, 30, 35, 40, 45, 50, 60, 70, 625. The median is the average of the fifth and sixth values, `(40 + 45) / 2 = 42.5`. The mean is `1000 / 10 = 100`. One household in ten sits above the mean; nine sit below it. The mean of 100 describes no household in this list — it is the per-household share of the total, not a typical income. Raise that last household from 625 to 1,225. The mean jumps from 100 to 160. The median does not move by one unit, because the fifth and sixth ranked values are still 40 and 45. That is the whole mechanism: **the mean is a function of the values, the median is a function of the ordering.** ## Stating the rule correctly The usable form is: for a unimodal distribution with a long right tail, mean > median, and typically mean > median > mode. Mirror it for a long left tail: mean < median. For a symmetric distribution the mean and median coincide (whenever the mean exists at all). Two qualifications an interviewer likes to hear: 1. It is a **heuristic, not a theorem**. Discrete and multimodal distributions can be constructed where the tail runs right yet the mean falls below the median. Nobody expects you to produce such a counterexample; they expect you not to claim the rule is a proof. 2. `mean = median` does **not** imply the data are normal, or even symmetric — it implies the imbalance in the two directions happens to cancel. Symmetry is the weaker claim that is usually meant, and normality is a much stronger claim requiring the whole shape. ## Which one to report The honest rule is to report the statistic that answers the question being asked. - **Typical experience** — what one household, one user, one order looks like: use the **median**. It is unaffected by how extreme the extremes are. - **Totals and budgets** — anything where the sum matters: use the **mean**, because `mean x n = total`. If you are sizing infrastructure spend or a payroll budget, the mean is the number that reconciles with the total, and the median simply does not. - **Diagnostics** — report both. The *gap* between them is itself information: a mean far above the median tells the reader the column is skewed before they see any chart. An optimisation characterisation makes the contrast precise. The mean is the constant c that minimises the sum of squared deviations `sum (x_i - c)^2`; the median is the constant that minimises the sum of absolute deviations `sum |x_i - c|`. Squaring is what makes far-away points expensive, and that is why the mean chases the tail while the median does not. ## Other centres in the same family For completeness, the arithmetic mean is not the only mean. A **weighted mean** assigns each value a weight and divides by the sum of the weights. A **trimmed mean** drops a fixed fraction from each end before averaging, landing between the mean and the median as a measure of centre. For quantities that multiply rather than add, the **geometric mean** is the appropriate average; for rates over a fixed numerator, the **harmonic mean** is. Knowing that the arithmetic mean is one choice among several — rather than the default definition of average — is what separates a mechanical answer from a fluent one. ## Answering it in an interview Say the mechanism first (magnitudes versus ranks), give a one-line numeric illustration, then name the reporting consequence: quote the median for a typical income, the mean when the total matters, and both when you are describing the shape.
- Which figure should a news headline quote for typical household income?The median. It answers what a household in the middle actually earns and is unmoved by how extreme the highest incomes are. The mean answers a different question — the per-household share of total income — and in a right-skewed column it sits above what most households experience, so quoting it as typical overstates the everyday reality.
- What does it tell you when a column's mean and median are almost identical?That the pull from each side roughly cancels, so the distribution is approximately symmetric around its centre. It is not evidence of normality: symmetric distributions can be flat, sharply peaked or heavy-tailed. Read it as one weak shape clue and confirm with a histogram before assuming anything stronger.
- In what sense is the median the value that minimises total distance to the data?The median is the constant c minimising the sum of absolute deviations, sum |x_i - c|. The mean is the constant minimising the sum of squared deviations, sum (x_i - c)^2. Squaring makes distant points disproportionately expensive, so the mean is pulled toward a long tail while the median, penalising distance linearly, is not.
- Would you ever report the mean for a right-skewed cost column?Yes, whenever the total matters. Mean multiplied by the number of records reproduces the total, so for budgeting, capacity planning or revenue reconciliation the mean is the only centre that adds up. Report it alongside the median so the reader knows the two differ and by how much.
The mean is the balance point of a see-saw loaded with the data; the median is the person standing in the middle of the queue. Move the heaviest person further out and the see-saw tips, but the queue order is unchanged.
saying these in an interview costs you the question
- Says the mean is always the best summary of a typical value
- Calls a long right tail left-skewed
- Claims mean equal to median proves the data are normal
- Thinks the median wastes information by ignoring half the data
- Believes doubling the largest value moves the median as much as the mean
- Treats a mean far above the median as evidence of a data-entry error