DataTerminology.comNEW
Also known as: MAD
A robust measure of variability computed as the median of absolute deviations from the median.
DESCRIPTION
Median Absolute Deviation (MAD) quantifies dispersion by taking the median of absolute differences from the dataset’s median, making it resistant to outliers. It is widely used in robust statistics, anomaly detection, and data quality checks because it does not inflate under heavy-tailed distributions. MAD complements standard deviation when data are skewed or contain extreme values. It is often scaled to approximate standard deviation under normal assumptions. In practice, MAD guides thresholding in outlier detection, robust normalization, and feature scaling. Its simplicity and interpretability make it a common diagnostic in exploratory data analysis.
ALTERNATIVE DEFINITIONS
NIST Engineering Statistics Handbook
The median absolute deviation (MAD) is defined as median(|Xi − median(X)|).
Provides the precise formula and emphasizes robustness to outliers.
Wolfram MathWorld
The median absolute deviation is the median of the absolute deviations from the median of a data set.
Highlights the median-based nature of the dispersion measure.
Statsmodels Documentation
Median absolute deviation of a sample.
Defines MAD in the context of a widely used Python library function.
RELATED TERMS
Variance Robust Statistics Data Anomaly Detection Median Quantile Z-score
COMPETING TERMS
Standard Deviation Interquartile Range Mean Absolute Deviation
KNOWLEDGE AREAS
Mathematics Data Science Testing & Validation Data Quality
TARGET AUDIENCE
Data Scientist Data Analyst Data Engineer Data Steward MLOps Engineer
EXTERNAL SOURCES
- Wikipedia - Median absolute deviation
- NIST Engineering Statistics Handbook - Official robust statistics reference
- Wolfram MathWorld - Mathematical definition and properties
- Statsmodels - Python implementation reference
- R Project - R function documentation for MAD
TOP QUESTIONS
Why is Median Absolute Deviation useful compared to Standard Deviation?
MAD is robust to outliers because it uses medians, not means. In heavy-tailed or skewed data, MAD provides a more stable variability estimate, improving anomaly detection and resilient analytics.
Category: Relevancy
How do you compute Median Absolute Deviation?
Compute the median of the data, subtract it from each value, take absolute deviations, and then take the median of those deviations. Optionally scale by a constant for normal distributions.
Category: Requirements
When should Median Absolute Deviation be preferred?
Use MAD when data contain outliers, are skewed, or have heavy tails. It offers reliable dispersion estimates for robust modeling, preprocessing, and outlier detection.
Category: Opinion
What is a common misunderstanding about Median Absolute Deviation?
Many assume MAD equals standard deviation. MAD measures median-based dispersion and is typically scaled to approximate standard deviation only under normal assumptions.
Category: Misunderstanding
How does Median Absolute Deviation support data quality?
MAD helps profile variability and flag anomalies without being distorted by extreme values. It enables more accurate thresholds and alerts in data quality monitoring.
Category: Management
What are risks of using Median Absolute Deviation incorrectly?
Using unscaled MAD for normal data can mislead comparisons with standard deviation. Misinterpreting MAD units or thresholds may cause false positives or negatives in detection.
Category: Risks
What are alternatives to Median Absolute Deviation?
Standard deviation, interquartile range, and mean absolute deviation are common alternatives. Choose based on robustness needs and distribution assumptions.
Category: Alternatives
How is Median Absolute Deviation applied in machine learning?
MAD informs robust feature scaling, outlier filtering, and residual analysis. It can improve model stability and reduce sensitivity to extreme observations.
Category: Example
Where can I learn more about Median Absolute Deviation?
Consult NIST’s handbook, MathWorld, and statistics packages like R and Statsmodels. Practice by computing MAD in exploratory data analysis workflows.
Category: Topic