Statistics › Section B · Descriptive statistics › Topic B.6

Skewness and the mean, median, and mode

Symmetric single-peaked data let the mean, median and mode agree; skew forces them into a predictable order. This topic turns that ordering into a tool - a numeric shape detector that works even when no graph is to hand.

Shape runs the centres

The mean, median and mode all measure the centre, but shape decides how they relate. In a symmetric, single-peaked distribution the three coincide: same value, same story. Skew pulls them apart in a predictable order, and the mean - because it counts every value - travels furthest. Learning the ordering turns three numbers into a shape detector: compute the centres, compare them, and you know which way the distribution leans before drawing anything.

Symmetric: the centres agree

A symmetric distribution has mirror halves around its centre, so its mean and median coincide. If it has a single peak, the mode sits there too; a symmetric bimodal distribution has two modes, one on each side of the shared mean and median. That agreement is the numeric signature of symmetry. The reverse works as a heuristic too: when a summary reports a mean and median close together, the underlying shape is probably roughly symmetric - not necessarily bell-shaped, just balanced left against right. Repeated measurements of the same quantity and the heights of adults of one sex often look this way.

Right skew: the mean chases the tail

A right-skewed distribution has its bulk at low values and a long tail toward high values. Incomes, waiting times and house prices are the classic cases: most values modest, a few enormous. The tail pulls the mean upward hardest, so the usual ordering runs mode, then median, then mean - for the TV hours of the centre topic (mean 3, median 2.5, mode 2), the mean sat highest. The rule is a guideline, not a law: with small or lumpy data the mode can misbehave, and a mean sitting above the median is the more useful right-skew signal.

Worked example: a left-skewed quiz

A left-skewed distribution is the mirror image: its bulk sits at high values and its tail runs toward low values. Easy-quiz scores are the classroom case - most students high, a few low. The tail pulls the mean downward, giving the usual ordering mean, then median, then mode. Work the scores 5, 8, 8, 8, 9, 9, 10, 10, 10, 10: the mean is \(\frac{87}{10} = 8.7\), the median is 9 and the mode is 10 - mean below median below mode, the left-skew signature. Check: the sum \(5+8+8+8+9+9+10+10+10+10 = 87\) over ten scores gives 8.7, and the two middle values are both 9. Again a guideline rather than a law; the mean-median comparison is the steadier part.

Why the mean moves most

The balance-point picture from the centre topic explains the whole ordering. The mean is the fulcrum that makes deviations cancel, so every value's distance counts: the few values in a long tail sit far out and shift the fulcrum toward themselves. The median counts only positions - which side of the middle a value falls on - so distant values can move house without changing the median's neighbourhood at all. The mode counts only the tallest pile. Hence the mean reflects skew most, the median moderately, and the mode least - the pattern the TV-hours and quiz examples display.

Worked example: three cafés, three shapes

Customer counts for eight hourly periods: Café A: 12, 15, 16, 18, 20, 24, 30, 65. Café B: 2, 20, 24, 26, 28, 29, 30, 31. Café C: 15, 18, 21, 24, 24, 27, 30, 33. 1. Means: A \(= \frac{200}{8} = 25\); B \(= \frac{190}{8} = 23.75\); C \(= \frac{192}{8} = 24\). 2. Medians: A \(= \frac{18+20}{2} = 19\); B \(= \frac{26+28}{2} = 27\); C \(= 24\). 3. Compare: A's mean sits above its median - right skew, driven by the 65-customer rush hour. B's mean sits below its median - left skew, from one dead 2-customer hour. C's mean equals its median - symmetric, matching values mirrored around 24. 4. Check: in every case the mean leaned toward the tail.

A numeric skew detector

Comparing the mean and median is the fastest numeric test for skew, and published summaries often give exactly those two numbers. Mean noticeably above median: suspect a right tail. Mean below median: suspect a left tail. Mean and median equal: suspect symmetry. The test is not proof - it says "go look at the shape" - but it flags the cases where reporting only the mean would mislead. It doubles as a reporting rule: when the detector fires, the median is the centre to quote, with the IQR beside it.

Skew is not the same as outliers

An outlier is one or a few points far from the rest; skew is the posture of a whole distribution - one entire side drawn out into a tail. The two often travel together, since one huge value right-skews a small data set, but neither requires the other: a smoothly right-skewed income distribution can contain no single outlier, and a symmetric data set can carry one bad typo. The mean-median detector responds to both, so when it fires, ask whether you are seeing a lopsided shape or a few stray points - the fixes differ. A stray point gets investigated; a real tail changes which centre you report.

Common mistakes to avoid

First, reading the skew direction from the peak: skew names where the tail points, not where the pile sits, and a right-skewed distribution usually peaks at its low end. Second, applying the three-centre ordering as a law - it holds only usually, and small data sets break it. Third, concluding symmetry from one pair of equal centres without checking a graph. Fourth, reporting the mean for visibly skewed data just because it is the familiar average. Fifth, deleting tail values to "fix" a skew: a real tail is data, not dirt.

Key takeaways

Symmetric data: mean and median agree, and so does the mode if there is one peak. Right-skewed data: the mean is pulled above the median (usually mode < median < mean) because the high tail tugs the balance point. Left-skewed data: the mean sinks below the median (usually mean < median < mode). The mean reflects skew most because it counts every value; the median counts only positions. Comparing mean with median is a quick numeric skew detector - and when it fires, report the median with the IQR, after checking whether the lopsidedness is a genuine tail or a few outliers.

Flashcards

Flashcard

Symmetric shape

Mean and median coincide

With one peak, the mode sits there too.

Test your understanding

Question 1 of 5

In a right-skewed distribution, the three centres usually appear in which order?

Question 2 of 5

A data set has mean 8.7 and median 9. Its shape is likely:

Question 3 of 5

In a right-skewed distribution, which centre typically sits closest to the long tail?

Question 4 of 5

A data set has mean = median = mode. Its shape is most likely:

Question 5 of 5

Café A's eight hourly customer counts have mean 25 and median 19. What does this comparison suggest?

← Previous topicNext topic →