Module 1: Describing data · Lesson 1.4 · about 25 min
Shape, skew and outliers: reading a box plot
Read the shape of data from a histogram and a box plot: symmetric or skewed, one peak or two, and which points are outliers by the 1.5 × IQR rule. What to do with an outlier once you find one.
Module 1 · about 25 minutes · school maths. You’ll be able to describe the shape of a distribution, read and draw a box plot, find outliers with the 1.5 × IQR rule and decide what to do with them.
In everyday terms: a box plot is like a summary of a race: where the middle runners crossed (the box), where most of the field finished (the whiskers), and the one runner who stopped for a picnic (the dot far out). You would ask the picnic runner what happened before deciding they don’t count.
1. Shape tells you which summary to trust
Two datasets can share a mean and SD and still look nothing alike. Before quoting any summary, look at the shape: draw a histogram (bars of how many values fall in each range) or a box plot. Ask four questions:
- Symmetric or skewed? A long tail to the right (order values, salaries, waits) is right skew or positive skew. A long tail to the left (exam scores when most do well, product ratings) is left skew.
- One peak or two? Two peaks (bimodal) usually mean two groups mixed together, such as weekday and weekend orders, or two store formats. Split them before summarising.
- Any gaps or walls? A wall at zero (no negative times) or at a limit (a 5-star cap) bends the shape.
- Any outliers? Values far away from the rest.
2. The box plot
A box plot compresses a distribution into five numbers plus outliers:
- the box from Q1 to Q3, with a line at the median;
- whiskers out to the furthest values still within 1.5 × IQR of the box;
- dots beyond the whiskers: outliers.
Stretch the tail and watch: the median sits off-centre in the box, the right whisker grows, the mean (dashed) moves above the median, and the skewness number turns positive. Add a 90-minute delivery and it appears as a separate dot.
Box plots shine when comparing groups side by side: delivery times per region, basket values per store. You see centre, spread, skew and outliers for each group at a glance.
3. The 1.5 × IQR rule
John Tukey’s rule of thumb marks a value as an outlier when it lies outside the fences
For normal data this flags only about 0.7 % of values. For skewed data it flags many values in the long tail, and that is fine: it is a prompt to look, not a verdict.
4. Skewness as a number
Skewness measures lopsidedness: 0 for symmetric data, positive for a right tail, negative for a left tail. As a rough guide, beyond ±1 is strongly skewed. Excel’s SKEW and pandas’ .skew() give the same adjusted formula used here.
5. What to do with an outlier
| Cause | Example | Action |
|---|---|---|
| Data error | Price typed as 1999 instead of 19.99 | Fix it, or remove it and note it |
| Different population | A business account in consumer orders | Analyse separately |
| Real, rare event | A lost parcel, a viral day | Keep it; report it; consider the median |
Never delete outliers just to make results look neater. Do say what you did: “3 test orders were removed; results with them included are in the appendix.”
- Box
- From Q1 to Q3: the middle half of the data, with a line at the median.
- Whiskers
- Reach to the furthest values still inside the fences.
- Fences
- Q1 − 1.5 × IQR and Q3 + 1.5 × IQR. Points beyond them are flagged as outliers.
- Right skew
- A long tail of large values; the mean sits above the median.
- Physical meaning
- Most business numbers have a floor (you cannot sell fewer than 0) and no ceiling, so they lean right: a few huge orders, long waits or big customers. The box plot shows that lean at a glance.
“Outliers are errors, so remove them.” Some are errors (a typo, a test order). Many are real and important: the biggest customer, the outage, the fraud. Find out which, and state what you did. Deleting real extremes makes risks look smaller than they are.
Watch it happen, one step at a time:
Maths and principle
Principle: the shape decides which summaries are honest. Symmetric data: mean and SD. Skewed data: median and IQR (or percentiles).
Equations: fences and . Sample skewness:
Cubing keeps the sign, so values far above the mean push it positive and far below push it negative.
Analogy: the box is the “normal range” of the data; the fences are the edge of the car park; anything outside gets a second look.
6. Worked example
Made-up numbers. Minutes to answer 11 support emails: 12, 15, 18, 20, 22, 24, 25, 27, 30, 34, 95.
- Median (6th value): 24. Q1 (interpolated, position 2.5 counting from 0): halfway from 18 to 20 = 19. Q3 (position 7.5): halfway from 27 to 30 = 28.5.
- IQR = 28.5 − 19 = 9.5; 1.5 × IQR = 14.25.
- Fences: 19 − 14.25 = 4.75 and 28.5 + 14.25 = 42.75.
- 95 is above 42.75: an outlier. The upper whisker stops at 34.
- Mean = 322 ÷ 11 ≈ 29.3, above the median of 24: right skew, mostly from the 95.
Before reporting, check the 95. If the email arrived on a Friday night and was answered Monday, it is real and matters for the service promise. If it was a test ticket, remove it.
7. Practice
- For Q1 = 40, Q3 = 70, find the outlier fences.
- A histogram of customer ages has peaks at 22 and 55. What does that suggest?
- Mean salary $68k, median $52k. Which way is the skew?
Answers
- IQR 30, 1.5 × 30 = 45. Fences −5 and 115.
- Two different customer groups (students and older customers, perhaps). Summarise each separately; a single average age of about 38 would describe almost nobody.
- Right skew: a few high salaries pull the mean above the median.
Try it on your own data: the box plot lab draws box plots from your numbers and marks outliers; the histogram builder shows the shape with any bin width.
New words: histogram, skew (right, left), bimodal, box plot, whisker, fence, outlier, skewness.
- Symmetric data: mean and SD work. Skewed data: prefer median and IQR.
- Box plot: box = Q1 to Q3, line = median, whiskers to the last values inside the fences.
- Outlier rule: below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR.
- Investigate every outlier: error, or a real extreme that matters?
Check yourself
Sources
- Tukey, J. W. (1977), Exploratory Data Analysis, Addison-Wesley (the box plot and fences)
- NIST/SEMATECH e-Handbook, 1.3.3.7 Box Plot https://www.itl.nist.gov/div898/handbook/eda/section3/boxplot.htm
Your ticks stay in this browser. Save them as a file on the course home.
Updated 2026-11-22. Numbers in the examples are made up to show the idea unless a source is given. Simulations run in your browser; nothing you type is sent anywhere.