Detect outliers using IQR (Tukey's hinges) and Z‑score methods with adjustable thresholds.Compute severity scores, visualize data distribution with a smart box plot, and export results to CSV.
An outlier is a data point that differs significantly from other observations in a dataset. Outliers can arise from measurement errors, data entry mistakes, or genuine variability in the population. Detecting and quantifying the severity of outliers is essential for data cleaning, robust statistical modeling, and ensuring the validity of analytical conclusions.
Severity Score = f( distance from central tendency, spread of the data )
Our calculator combines IQR and Z‑score methods to assign a normalized severity score from 0 (not an outlier) to 10 (extreme outlier).
Not all outliers are equally influential. A point just beyond the 1.5×IQR threshold has a different impact on statistical analyses than a point that is 10 standard deviations from the mean. Severity scoring helps prioritize which outliers warrant further investigation. In quality control, a severe outlier may indicate a critical process failure. In finance, extreme outliers can signal fraud or market anomalies. In healthcare, they may represent rare but clinically significant events.
The concept of outlier severity builds on foundational work by Tukey (1977) on exploratory data analysis, and later extensions by Rousseeuw & Leroy (1987) on robust regression and outlier diagnostics. Modern machine learning pipelines increasingly incorporate outlier severity as a feature for anomaly detection systems.
For each point identified as a potential outlier, we compute:
The final Severity Score is the maximum of the two, capped at 10. Points that are not flagged as outliers receive a severity score of 0. Severity levels are assigned as: Low (1–3) Moderate (4–6) High (7–8) Extreme (9–10).
| Method | Threshold | Best used for | Limitations |
|---|---|---|---|
| IQR (k=1.5) | Q1−1.5×IQR & Q3+1.5×IQR | Skewed data, non‑normal distributions | May miss outliers in small samples |
| IQR (k=3.0) | Q1−3×IQR & Q3+3×IQR | Extreme outlier detection | Less sensitive to moderate outliers |
| Z‑score (|z|>3) | |z| > 3 | Approximately normal data | Sensitive to outliers in mean/std |
| Z‑score (|z|>2.5) | |z| > 2.5 | Smaller samples, more sensitive | Higher false positive rate |
| MAD (median absolute deviation) | |x−median|/MAD > 3.5 | Robust to heavy tails | Less commonly used |
| Grubbs' test | p < 0.05 | Single outlier detection | Assumes normality, one outlier at a time |
A production line measures the diameter of precision bearings (in mm). The target diameter is 10.00 mm. A sample of 50 bearings yields a mean of 10.02 mm with a standard deviation of 0.05 mm. Two measurements are recorded as 10.35 mm and 9.68 mm. Using our outlier severity calculator:
The production team traced the first outlier to a worn tool bit, and the second to a mis‑calibrated gauge. By quantifying severity, the team prioritized corrective actions, reducing scrap rate by 18% in the following month.
In robust statistics, the influence function measures how much a statistic is affected by a single observation. Outliers with high severity have a large influence on the mean and variance, but less on the median and IQR. This is why robust estimators (like the median and MAD) are often preferred when outliers are expected. Our severity score helps you decide whether to exclude, transform, or keep an outlier in your analysis — a decision that should be based on domain knowledge and the severity of the outlier.
For further reading, we recommend Huber (1981) "Robust Statistics" and Wilcox (2012) "Introduction to Robust Estimation and Hypothesis Testing". These texts provide a rigorous foundation for understanding the impact of outliers on statistical inference.