Robust Descriptive Statistics
Also known as: resistant statistics, outlier-resistant summary statistics, robust summary measures, robust location and scale estimation
Robust descriptive statistics summarize the location, spread, and shape of a dataset using measures that remain meaningful even when a fraction of the data contains outliers or severe departures from normality. Core tools include the median, trimmed mean, interquartile range (IQR), and median absolute deviation (MAD), all of which are resistant to contamination that would distort the classic mean and standard deviation.
Key highlights
- Resistant to the distorting effect of outliers and heavy-tailed distributions.
- Median and MAD have a 50% breakdown point, the highest theoretically achievable.
- Provide an honest picture of the bulk of the data without requiring removal of observations.
- Easily communicated: medians and IQRs are intuitive to non-statistician audiences.
- Compatible with formal robust inference (robust t-tests, M-estimators) as a coherent analytic framework.
- Scale-independent: MAD and IQR preserve the original measurement units.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use robust descriptive statistics whenever normality cannot be assumed, sample sizes are small, or visual inspection reveals outliers or heavy-tailed distributions. They are especially appropriate for skewed data such as reaction times, income, or biological assays, and for exploratory work before selecting inferential tests. Do not substitute robust summaries for classical ones when the distribution is genuinely symmetric and outlier-free, since in that case the mean and standard deviation are more efficient and universally expected by reviewers. Also avoid relying solely on robust measures when the contaminating observations carry substantive scientific meaning — report and investigate outliers, rather than simply suppressing them.
Strengths & limitations
- Resistant to the distorting effect of outliers and heavy-tailed distributions.
- Median and MAD have a 50% breakdown point, the highest theoretically achievable.
- Provide an honest picture of the bulk of the data without requiring removal of observations.
- Easily communicated: medians and IQRs are intuitive to non-statistician audiences.
- Compatible with formal robust inference (robust t-tests, M-estimators) as a coherent analytic framework.
- Scale-independent: MAD and IQR preserve the original measurement units.
- Less efficient than the mean and standard deviation when the data are truly normally distributed.
- Trimmed means and Winsorized means require an a priori choice of trim proportion, which is somewhat arbitrary.
- Some journals and disciplines still require reporting classical means and SDs; robust measures may need to accompany, not replace, them.
- Do not eliminate the need to understand why outliers exist — robust measures mask rather than explain anomalous observations.
- MAD requires an additional consistency correction factor under normality, and its sampling distribution is less familiar to practitioners.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Should I always report both the mean and the median?
When the distribution is symmetric and free of outliers, the mean is the preferred location measure and the median adds little. When skew or outliers are present, the median more accurately represents the typical value. In practice, reporting both is safest for transparency: it signals to readers how much the distribution departs from symmetry.
What trim percentage should I use for a trimmed mean?
A 20% symmetric trim (removing the bottom and top 20%) is a common default that balances robustness and efficiency. In domains with known heavy contamination, higher trim rates may be justified. The proportion should be chosen before data inspection, not tuned to maximize a result.
Is the IQR the same as the standard deviation?
No. The IQR covers the middle 50% of the data and is a robust spread measure; the standard deviation averages squared deviations from the mean and is sensitive to outliers. Under normality, IQR ≈ 1.35 × SD. Both are useful but serve different purposes and should not be substituted for each other without acknowledgment.
Can I use robust descriptive statistics and then run a standard t-test?
Robust descriptive summaries describe the data; the choice of inferential test depends on assumptions separately evaluated. If the data require robust description (outliers, non-normality), a robust inferential test — such as the Welch-corrected t-test, Mann-Whitney U test, or Yuen's trimmed-mean test — is generally more consistent than pairing robust description with a classical parametric test.
Does using the median violate any reporting standards?
No. APA Publication Manual (7th ed.) explicitly recommends reporting the median when distributions are skewed and requires reporting a measure of spread alongside any location measure. Many medical journals similarly prefer median and IQR for non-normally distributed continuous outcomes.
Sources
- 1.Tukey, J. W. (1977). Exploratory Data Analysis. Addison-Wesley.ISBN 978-0201076165
- 2.Huber, P. J., & Ronchetti, E. M. (2009). Robust Statistics (2nd ed.). Wiley.ISBN 978-0470129906
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Robust Descriptive Statistics. ScholarGate. https://scholargate.app/statistics/robust-descriptive-statistics