Outlier detection (explained like you’re teaching it)

An outlier is a value that looks unusually far away from the rest of the data. Outliers can happen for boring reasons (a typo, a sensor glitch, a “comma in the wrong place”) or for interesting reasons (a true rare event, a breakthrough performance, a fraud spike, a once-in-a-decade storm). This is why outlier detection is used everywhere: statistics homework, machine learning, quality control, finance, healthcare, sports analytics, and basically any “numbers-based” work.

This calculator gives you three popular methods. They’re all trying to answer the same question: “How far is too far?” The difference is what they consider “typical” and how sensitive they are when your dataset is skewed, has heavy tails, or is small.

The IQR (interquartile range) is the spread of the middle 50% of your data. First, we compute the first quartile Q1 (25th percentile) and the third quartile Q3 (75th percentile). Then:

• IQR = Q3 − Q1
• Lower fence = Q1 − k × IQR
• Upper fence = Q3 + k × IQR

Any value below the lower fence or above the upper fence is flagged as an outlier. The slider k controls strictness. k = 1.5 is the classic default. k = 3 is often used for “extreme” outliers. IQR is popular because it doesn’t get pushed around as much by extreme values (it uses quartiles instead of the mean).

Z-score compares each value to the mean in units of standard deviation:

• z = (x − mean) / standard deviation
• Flag as outlier if |z| > threshold

The default threshold is often 3.0 because, for a perfectly normal (bell-curve) distribution, almost all values fall within ±3 standard deviations. If your data is roughly symmetric and bell-shaped, Z-score is fast and intuitive. But if your dataset is skewed (like salaries) or has heavy tails, the mean and standard deviation can be distorted by extreme values—making Z-score less reliable.

Modified Z-score is like Z-score, but it swaps in robust statistics: the median (instead of mean) and the MAD (median absolute deviation, instead of standard deviation). It’s a favorite when you want something that “doesn’t panic” when there’s a huge spike.

• MAD = median( |x − median| )
• modified z = 0.6745 × (x − median) / MAD
• Flag as outlier if |modified z| > threshold (often 3.5)

The constant 0.6745 scales MAD so that for normal data, the modified Z-score behaves similarly to the standard Z-score. If you’ve got messy real-world data (human behavior, spending, time-on-task, latencies), this method is often the safest first try.

• Data size (n): How many valid numbers you entered.
• Mean & standard deviation: Useful for Z-score and general summaries.
• Median, Q1, Q3, IQR: Useful for IQR fences and robust summaries.
• MAD: Useful for modified Z-score.
• Outliers list: Which values are flagged, plus the rule/fences used.

One final reality check: outlier detection is not a moral judgment. It’s a rule. A good workflow is: flag → investigate → decide. If you remove outliers, document the rule and the reason. If you keep them, explain why they’re real and meaningful.

Outlier examples you can copy

Data: 10, 12, 12, 13, 14, 40

• IQR will likely flag 40 as an outlier because it’s far above the upper fence.
• Z-score may also flag 40 depending on how large the standard deviation becomes.
• MAD usually flags 40 strongly because median-based spread stays small.

Data: 5, 6, 6, 7, 7, 7, 8, 9, 25

This dataset is right-skewed (one large value). IQR and MAD typically do well here. Z-score can be “too forgiving” because the mean and standard deviation get pulled upward.

Data: 98, 101, 102, 99, 100, 500, 510, 520

Are the 500+ values outliers? Maybe—but it might also be two different populations (e.g., two machines, two customer types, or two measurement scales). In that case, the best fix isn’t “remove them,” it’s “split the dataset.”

Data: 10, 10, 10, 10

Standard deviation and MAD become 0. If all values are identical, nothing can be an outlier (because there’s no spread). The calculator will show a friendly note if a method can’t run.

Which method should you use?

If you’re unsure, here’s a quick decision guide:

• Use IQR when your data is skewed, has weird spikes, or you need a simple “quartile fence” explanation.
• Use Z-score when your data is roughly normal (bell-shaped) and you want a classic stats approach.
• Use MAD when you suspect extreme outliers might distort the mean/standard deviation.

In many real projects, people run two methods and compare. If a value is flagged by both IQR and MAD, it’s probably genuinely extreme. If only one method flags it, that’s a signal to investigate rather than delete.

Paste your daily screen time, sleep hours, or spending amounts for the last month. The outlier you find is often the “story”: the one day you pulled an all-nighter, the surprise expense, or the day you accidentally walked 25,000 steps. Screenshot it and share it. 📸

Educational note: Quartile calculation conventions can differ (inclusive/exclusive methods). This calculator uses a common linear-interpolation percentile approach for stable results across sample sizes.

Frequently Asked Questions

• What counts as an outlier?

  An outlier is a value that falls outside a rule-based “expected” range. The IQR method uses quartiles, Z-score uses standard deviations from the mean, and MAD uses deviations from the median. Different rules can flag different values, especially in skewed datasets.

• Should I remove outliers?

  Not automatically. First ask why they exist: error, rare event, different population, or real but extreme. If you remove them, keep a note of your rule (e.g., “Removed values outside Q1 − 1.5×IQR and Q3 + 1.5×IQR”).

• What is the best threshold?

  For IQR, k = 1.5 is the classic standard; k = 3 is stricter. For Z-score, 3 is common; 2 is more sensitive. For modified Z-score, 3.5 is a popular default. The “best” threshold depends on how expensive false alarms are vs missed outliers.

• Why does Z-score sometimes miss obvious outliers?

  Because extreme values inflate the mean and the standard deviation. That can reduce the Z-score of the extreme value itself. If this happens, try MAD or IQR.

• What if standard deviation or MAD is zero?

  If all values are identical (or nearly identical), there’s no spread. In that case, a distance-based method can’t separate “typical” from “extreme.” The tool will tell you if it can’t compute Z-score or modified Z-score.

• Does the order of numbers matter?

  No. The calculator sorts internally for median/quartiles and uses the full list for mean and deviation.

• Can I paste values with commas (like 1,000)?

  The safest way is to paste plain numbers. If you paste “1,000” the comma may be treated as a separator. For thousand-separated numbers, remove the commas or paste as “1000”.