Concept · C:outlier

Outlier

Working definition

An observation that appears unusually distant or inconsistent under a declared distributional, graphical, or rule-based comparison and therefore warrants investigation.

Also calledUnusual observation · Potential outlier

Under the declared median-of-halves convention, Juniper's 16 Year 2 values are split into ordered lower and upper halves of eight. The lower half's middle pair is 22 and 23, so (Q_1=22.5); the upper half's middle pair is 30 and 31, so (Q_3=30.5). The interquartile range is 8 days. The upper inner fence is:

[ 30.5+1.5(8)=42.5\text{ days}. ]

The 78-day invoice lies beyond that fence, so the rule flags it. The rule does not say why it is there.

Investigate the record, not the preferred answer

Trace the invoice to issue, due-date, dispute, payment, cash-application, and adjustment evidence. Check the unit and key, data entry, customer terms, partial payments, and definition of final settlement. Then choose a documented action:

  • correct a verified recording or coding error;
  • retain a valid observation in the primary analysis;
  • report a robust or segmented analysis alongside it; or
  • exclude it only when a predeclared population or data-quality rule supports exclusion.

Deleting 78 because it widens a confidence interval or weakens a preferred story is result-driven analysis. Automatically retaining a known duplicate or wrong-date record is no better.

Quartile conventions vary across software. The fence is a transparent screening convention, not a universal classification or proof that the data follow a particular probability model.

Learning objectives

Put the concept to work

Learning level

Understand this concept

  • Explain why an outlier flag identifies an observation for investigation but does not by itself prove error, fraud, irrelevance, or a different population.
Learning level

Analyze this concept

  • Investigate a flagged business observation against source evidence and document correction, retention, separate modeling, or exclusion without result-driven deletion.

Learning resources

Choose a lesson, try an application, or inspect the sources behind this concept.

Build on these ideas

  • Distribution — Understand

    To understand this concept: Required. Unusualness is judged relative to a distribution and declared method.

  • Observational unit — Analyze

    To analyze this concept: Helpful. Duplicate and wrong-grain records are possible sources of extreme observations.

  • Outlier — Understand

    To analyze this concept: Required. A defensible treatment begins by separating an analytical flag from an error conclusion.

Lessons

Worked examples and cases

Show 2 more examples and cases

Practice

Common mistaken ideas

Sources

Show 1 more related concepts

Use this idea next

  • Outlier — Analyze

    Required level here: understand. Required. A defensible treatment begins by separating an analytical flag from an error conclusion.

Updated Aug 7, 2026 Review due Nov 7, 2026