This outlier calculator screens a dataset with the Tukey IQR method — quartiles, inner fences at 1.5 × IQR, and the specific values flagged as outliers (and extreme outliers beyond 3 × IQR).
An outlier flag is an investigation prompt, not a deletion order: data errors get corrected, genuine extremes get modeled (they’re often the point), and only documented mistakes get removed — with the removal disclosed. Deleting inconvenient points to prettify a fit is how analyses become fiction.
The Tukey Fences: Q1 − 1.5·IQR and Q3 + 1.5·IQR
Quartiles come from the median-of-halves method detailed on our Quartile Calculator. The 1.5 multiplier is Tukey’s convention — for normal data it flags about 0.7% of points — and the 3× fences mark “extreme” outliers. The method’s virtue is robustness: the fences are computed from quartiles, which the outliers themselves barely move (unlike mean ± SD rules, where the suspect points inflate the very yardstick judging them). Part of our Statistics Calculators collection.
Worked Example (Verified)
- 1–10 plus 100 (defaults): halves medians give Q1 = 3, Q3 = 9, IQR = 6 → fences −6 to 18 → 100 flagged, and it’s beyond the 3× fence (27) too: extreme ✔
- The masking contrast: same data judged by mean ± 2SD — mean 14.1, SD 28.5 — the “fence” reaches 71 and even 100 barely convicts; IQR’s robustness is structural, not luck
- Clean data: 1–10 alone → Q1 3, Q3 8, fences −4.5–15.5 → none flagged ✔ (a screen that cries wolf on tidy data would be useless)
- Symmetric heavy tails break the spirit, not the math: genuinely skewed data (incomes) flags its entire upper class — transform (log) before judging, or use a skew-aware method; a flag rate of 15% indicts the method-data match, not the data
What To Do With a Flag
- Trace it: transcription error? unit mix-up (cm among inches)? sensor glitch? — provenance first, statistics second.
- Keep genuine extremes and choose methods that tolerate them (medians, robust regression), or model them explicitly.
- If you remove points, report how many and why — silent deletion is the reproducibility crisis in miniature.
- Small samples (n < ~8) make fences unreliable; the method is a screen for moderate data, not a verdict for tiny sets.
Related Tools
- Quartile Calculator — Q1, median, Q3 in detail
- Standard Deviation Calculator — the non-robust yardstick
- Statistics Calculators — the full collection
Frequently Asked Questions
How do you find outliers with the IQR method?
Compute Q1, Q3 and IQR = Q3 − Q1. Flag anything below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR (Tukey fences); beyond 3 × IQR counts as extreme. For 1–10 with 100 added: Q1 3, Q3 9, fences −6 to 18 — only 100 is flagged.
Why 1.5 × IQR?
Tukey’s chosen convention (1977) — pragmatic, not derived from a theorem. On perfectly normal data it flags ≈ 0.7% of observations, a rate generations of analysts have found balances sensitivity against noise. The 3× “far out” fence is its stricter sibling.
Should I delete outliers?
Only documented errors, and transparently. Genuine extreme values carry information — fraud signals, rare diseases, record performances — and deleting them biases every downstream statistic toward a tidier fiction. Prefer robust summaries (median/IQR) that coexist with the extremes.
Outlier vs anomaly vs error — same thing?
No. An outlier is a statistical flag (unusual relative to a rule). An anomaly is a substantive judgment (doesn’t belong to the process). An error is a provenance fact (recorded wrong). The workflow runs flag → investigate → decide; skipping the middle step conflates all three.
Which quartile method does this use?
Tukey’s median-of-halves (hinges): split the sorted data at the median; Q1 and Q3 are the halves’ medians. Other conventions (inclusive percentile methods) can shift quartiles slightly on small datasets — links above show the Quartile Calculator’s detailed breakdown; conclusions rarely hinge on the difference at moderate n.
Is this outlier calculator free?
Yes — free, instant, no sign-up.