About the Modified Z-Score Outliers
This calculator flags outliers in a dataset using the modified z-score method, a robust alternative to the standard z-score that substitutes the median and median absolute deviation for the mean and standard deviation. It suits datasets where a handful of extreme values could otherwise corrupt the very statistics used to detect them.
How It Works
Enter your numbers and an outlier threshold (3.5 by default). The calculator finds the median and the median absolute deviation (MAD) of the data, then converts every value into a modified z-score. Any value whose modified z-score exceeds the threshold in absolute value gets listed as an outlier.
Formula & Methodology
The constant 0.6745 is the 75th percentile point of the standard normal distribution; it rescales MAD so that, for normally distributed data, the modified z-score behaves on roughly the same scale as an ordinary z-score. Compute the median of all values, then the MAD (median of absolute deviations from that median), then apply the formula to each point.
Examples
Default dataset with one clear outlier
For 4, 6, 7, 7, 8, 9, 10, 45 (median 7.5, MAD 1.5), the value 45 has a modified z-score of 0.6745 * (45 - 7.5) / 1.5 = 16.86, far past the 3.5 threshold, while every other value stays under 1.6 in absolute value, so only 45 is flagged.
Small dataset, threshold reached
For 2, 3, 3, 4, 5, 7, 10 (median 4, MAD 1), the value 10 scores 0.6745 * (10 - 4) / 1 = 4.05, above the 3.5 threshold and flagged, while 2 scores -1.35 and stays well within range.
Advantages
- Substitutes median and MAD for mean and standard deviation, so the detection threshold itself isn't skewed by the outliers under investigation.
- Produces a single adjustable threshold, letting users tighten or loosen sensitivity for their specific dataset instead of relying on a fixed rule.
- Reports the underlying median and MAD alongside the outlier list, so the basis for each flag is visible, not just the final verdict.
Common Mistakes
- Applying the default 3.5 threshold to every dataset without considering that a stricter or looser cutoff may fit the data's context better.
- Running the test on fewer than 3 data points, where the median and MAD carry too little information to meaningfully flag anything.
- Assuming a flagged "outlier" is automatically an error or bad data point, when it may simply be a genuine, rare extreme value worth investigating rather than discarding.
Edge Cases to Watch For
- Requires at least 3 numbers; fewer than that returns an error.
- If MAD comes out to exactly 0 (common when more than half the values repeat), the formula would divide by zero, so the calculator returns an error instead of a result.
- The default threshold of 3.5 follows Iglewicz and Hoaglin's commonly cited recommendation, but the field is adjustable: a lower threshold flags more values, a higher one flags fewer.
- Because it uses absolute value, both unusually high and unusually low values can be flagged; a single-sided extreme dataset will only ever flag one direction.
Common Use Cases
- Data cleaning workflows that need to flag suspicious values in skewed or small datasets before further analysis.
- Quality control monitoring where a handful of measurements might be corrupted by sensor error or manual entry mistakes.
- Financial or operational analysts scanning for unusual transactions or metrics in datasets where a standard z-score would be thrown off by the outliers themselves.