Skip to content

MODULE 3 · LESSON 16 · Essential statistics

Distributions and outliers

Describe a distribution and investigate unusual observations without automatically deleting them.

By Laurence Svekis · Self-paced · All worked data is synthetic unless you supply your own source.

Try the practice ↓

PredictInspectCalculateCheckExplainTransfer

Your starting point

Prerequisites: Use the ideas and vocabulary from lesson 15; revisit its worked examples if needed.

Definition: A distribution describes how values are arranged across their possible range. An outlier is an observation unusually far from others under a stated rule or context. A boxplot-style fence often flags values below Q1−1.5×IQR or above Q3+1.5×IQR, where IQR = Q3−Q1.

Look beyond a single summary

Inspect the center, spread, asymmetry, clusters and gaps. A long right tail means a few large values extend the upper side. Two clusters may indicate different populations or processes. A histogram can reveal these patterns, but its bins influence what appears visible. Compare multiple reasonable bin choices without selecting only the one that supports your preferred story.

A flag is a question, not a verdict

An extreme value may be a typo, a unit mismatch, a rare real event or evidence of a different process. Check the source and context. Deleting a real long delay can make a service look better while hiding the case most important to users. If you exclude a confirmed error, document the evidence and compare the result before and after correction.

State the rule you used

Different software can compute quartiles differently on small samples. The academy uses linear interpolation at position (n−1)p in the ordered values. Its fences are exploratory flags rather than formal hypothesis tests. A point outside a fence is not proof of fraud or an impossible measurement. Consider the subject matter and the observation process alongside the numerical rule.

Two worked examples

An exploratory fence

For 1, 2, 2, 3 and 100, the academy’s quartiles are Q1 = 2 and Q3 = 3. IQR is 1, so the upper fence is 4.5. The value 100 is flagged. The calculation identifies an unusual record; it does not establish whether 100 is a typo or a genuine event.

Two legitimate groups

Response times 2, 3, 4, 20, 21 and 22 may come from chat and email channels. A single overall mean hides two clusters. Separate by channel and inspect the collection process before describing the combined distribution as one homogeneous service.

Where this becomes useful

An operations analyst sees a single very large order. It could be an important wholesale customer rather than a data error. The decision to retain it, segment it or correct it depends on the question and source evidence.

Make an attempt

Use Distribution Visualizer with 1,2,2,3,100. Inspect the fence, then change the number of bins. Keep the values fixed and describe which visual features change. Record a source check you would need before removing 100.

Open the accompanying interactive lab →

Misconception check

Claim: Outliers are just bad data.

Inspect the correction

Some are errors; others are real and consequential. Investigate rather than removing inconvenient observations.

Independent application

Analyze a small distribution with one unusual value. Document the rule used to flag it, compare summaries with and without it, and write two conditional interpretations: one if it is a confirmed error and one if it is a valid rare event.

Completion records practice and self-review. It is not certification of statistical competence.