Breakdown Point Analysis
Also known as: breakdown point, finite-sample breakdown point, robustness breakdown analysis, Bozunma Noktası Analizi
Breakdown point analysis quantifies the fraction of outliers an estimator can tolerate before it produces meaningless results. Formalised by Hampel (1971) and Donoho and Huber (1983), it is the standard tool for comparing the robustness of competing estimators.
Key highlights
- Gives a single, interpretable number for how many outliers an estimator can tolerate.
- Enables direct, comparative robustness assessment across competing estimators.
- Grounded in well-established robustness theory from Hampel and from Donoho and Huber.
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 breakdown point analysis when you need to choose or justify an estimator on the basis of its resistance to outliers and contaminated data, with at least about 10 observations. It applies to continuous data and is most informative when a specific estimator (such as OLS, LTS, or an M-estimator) has been identified for evaluation. Keep in mind that the breakdown point can never exceed 0.5, and that a high breakdown point typically comes at the cost of statistical efficiency. It is less meaningful for very small samples, where convergence is slow and the measure becomes statistically unstable.
Strengths & limitations
- Gives a single, interpretable number for how many outliers an estimator can tolerate.
- Enables direct, comparative robustness assessment across competing estimators.
- Grounded in well-established robustness theory from Hampel and from Donoho and Huber.
- The breakdown point is bounded above by 0.5; no estimator can survive more than half the data being contaminated.
- A high breakdown point usually trades off against reduced statistical efficiency.
- It is statistically meaningless for very small samples (n < 10), where the estimator should be chosen directly.
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
What does a breakdown point of 0.5 mean?
It means the estimator can tolerate up to half of the data being replaced by arbitrary values before its estimate can be driven to infinity. This is the theoretical maximum: no equivariant estimator can have a breakdown point above 0.5, because beyond that point the contaminated and clean halves become indistinguishable.
Why is the breakdown point of OLS so low?
Ordinary least squares has a breakdown point of 1/n, which tends to 0 as the sample grows. A single sufficiently extreme observation can pull the OLS fit arbitrarily far, which is why it is considered non-robust and why robust alternatives are needed under contamination.
Does a higher breakdown point always make an estimator better?
No. A higher breakdown point usually comes at the cost of lower statistical efficiency, meaning less precise estimates on clean data. The analysis is about balancing resistance to outliers against efficiency, not about maximising robustness alone.
How small a sample is too small for this analysis?
Below about 10 observations the breakdown point is statistically meaningless and you should select the estimator directly; below 20 its convergence is slow, so the results need to be interpreted with care.
Sources
- 1.Donoho, D. L. & Huber, P. J. (1983). The Notion of Breakdown Point. In A Festschrift for Erich L. Lehmann (pp. 157-184). Wadsworth.
- 2.Hampel, F. R. (1971). A General Qualitative Definition of Robustness. Annals of Mathematical Statistics, 42(6), 1887-1896.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Breakdown Point Analysis. ScholarGate. https://scholargate.app/statistics/breakdown-point-analysis