Skip to content

Median Absolute Deviation

A robust measure of univariate dispersion defined as the median of the absolute distances from the sample median, resistant to a minority of extreme observations.

Version
v1 · 2026-09-28 · History
Domain-specific #
10649
Domain group
Formal Sciences
Origin domain
Experimental Design & Statistics
Subdomain
Robust Statistics → Experimental Design & Statistics

Core Idea

The median absolute deviation (MAD) is a robust scale statistic for one-dimensional quantitative data. First find the sample median, take every absolute distance from that median, and then take the median of those distances.

For the data 1, 1, 2, 2, 4, 6, 9, the center is 2 and the absolute deviations are 1, 1, 0, 0, 2, 4, 7, whose median is 1. Extreme observations have limited effect because neither the center nor the final aggregation is a mean of squared distances. Some applications multiply raw MAD by a consistency factor, commonly for normal data, so formula and scaling convention must accompany reported values.

Structural Signature

Sig role-phrases:

  • univariate quantitative sample. Supplies observations on one comparable scale. Constitutive data. If altered: Multivariate dispersion needs another construction.
  • sample median. Provides the robust center. Identity-bearing reference. If altered: Using the mean defines mean absolute deviation instead.
  • absolute deviations. Measure unsigned distances from the median. Constitutive transformation. If altered: Squared deviations define variance-related measures.
  • median aggregation. Selects the middle transformed distance. Constitutive robust summary. If altered: A minority of arbitrarily large distances does not dominate it.
  • scale convention. States raw MAD or a consistency-scaled version and handles even samples or zero values. Necessary interpretation boundary. If altered: Software may multiply by a normal-consistency constant.

What It Is Not

  • Mean absolute deviation. Is averaging used?
  • Standard deviation. Are deviations squared and mean-centered?
  • Interquartile range. Are quantile positions rather than center distances used?
  • Scaled MAD. Was a constant applied to raw MAD?

Scope of Application

Use MAD with data units, missing-value treatment, sample/population meaning, raw or scaled convention, tie handling, and intended distribution stated.

  • Robust statistics. Estimates scale under contamination.
  • Outlier detection. Builds robust standardized distances.
  • Quality control. Monitors skewed or heavy-tailed data.
  • Signal processing. Summarizes noise amplitude.
  • Exploratory analysis. Compares spread with medians.

Clarity

MAD can be zero when at least half the observations equal the median, even though other values vary.

Manages Complexity

Robustness does not mean universal efficiency or adequacy. Discrete, asymmetric, multimodal, censored, or very small samples can need different scale descriptions and interval methods.

Abstract Reasoning

  1. Confirm one quantitative scale and clean missing values.
  2. Compute the sample median.
  3. Form absolute deviations from that median.
  4. Take their median and retain units.
  5. Declare any scaling factor and interpret distribution limits.

Knowledge Transfer

Median-centered robust distance transfers to residual analysis, but univariate data and the exact two-stage median operation delimit MAD. The nearest stopping boundary is explicit: Mean absolute deviation from the median is closest: it uses the same distances but averages them, losing the defining median aggregation and some robustness. The inclusion test remains: A statistic is MAD when it takes the median of absolute deviations from the sample median under a declared scaling convention. The structure no longer applies when the case exits when deviations are centered elsewhere or summarized by something other than their median.

Examples

Canonical

For 1,1,2,2,4,6,9, median 2 yields deviations 1,1,0,0,2,4,7 and raw MAD 1.

Mapped back: univariate quantitative sample → seven values; sample median → 2; absolute deviations → listed distances; median aggregation → 1; scale convention → raw MAD.

Applied / In Practice

A program averages absolute distances from the mean and labels the result MAD. That is mean absolute deviation, not median absolute deviation from the median.

Mapped back: univariate quantitative sample → present; sample median → not used; absolute deviations → from mean; median aggregation → replaced by average; scale convention → different statistic.

Structural Tensions

T1: outlier resistance vs. efficiency. Robustness protects against extremes but can sacrifice precision in ideal light-tailed models. Diagnostic: What contamination is expected?

T2: simple formula vs. software convention. Scaled and raw outputs can share the same name. Diagnostic: Was a consistency constant applied?

Structural–Framed Character

Description turns on univariate quantitative sample, sample median, absolute deviations, median aggregation, scale convention. Skeletal core. A robust center anchors distances that are themselves robustly aggregated. Domain-bound accent. Samples, medians, absolute deviations, outliers, scale factors, and dispersion define MAD. Transfer remains bounded because Why not prime. Robust distance summarization is portable; this is a named statistic. The negative boundary is concrete: Any mean absolute deviation, standard deviation, variance, median error, interquartile range, absolute residual, robust z-score, scaled MAD, or maximum deviation is not automatically the raw median absolute deviation. MAD is statistical-measurement: two median operations produce a robust estimate of typical absolute spread. Its character: variability summarized by the middle distance from the middle value.

Structural Core vs. Domain Accent

Skeletal core. A robust center anchors distances that are themselves robustly aggregated.

Domain-bound accent. Samples, medians, absolute deviations, outliers, scale factors, and dispersion define MAD.

Why not prime. Robust distance summarization is portable; this is a named statistic.

  • Median. It supplies both center and aggregation.
  • Robust scale. MAD resists a minority of extremes.
  • No strict parent is asserted.

Neighborhood in Abstraction Space

Median Absolute Deviation sits in a moderately populated region (45th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Empirical Measurement & Statistical Inference Methods (50 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With

  • Mean absolute deviation. Tell: Is averaging used?
  • Standard deviation. Tell: Are deviations squared and mean-centered?
  • Interquartile range. Tell: Are quantile positions rather than center distances used?
  • Scaled MAD. Tell: Was a constant applied to raw MAD?

References

  • Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Median_absolute_deviation (revision 1345734858).
  • Preserved source candidate: https://books.google.com/books?id=i2bD50PbIikC&pg=PA118
  • Preserved source candidate: https://dipot.ulb.ac.be/dspace/bitstream/2013/139499/1/Leys_MAD_final-libre.pdf
  • Preserved source candidate: https://crates.io/crates/rstats
  • Preserved source candidate: https://archive.org/details/operationsmanage0005russ/page/497

The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.