Robust Control Chart — Outlier-Resistant Statistical Process Monitoring
Also known as: robust Shewhart chart, outlier-resistant control chart, robust SPC chart, distribution-free control chart
A robust control chart replaces the classical mean and standard deviation estimators in a Shewhart-style chart with resistant alternatives — such as the median and median absolute deviation (MAD) — so that a small fraction of outliers or non-normal process data cannot distort the control limits. The approach preserves the real-time monitoring logic of standard control charts while protecting against inflated or deflated limits caused by contaminated Phase I reference data.
Key highlights
- Produces reliable control limits even when Phase I baseline data contain a moderate fraction of outliers or special-cause events.
- Inherits the real-time monitoring simplicity and interpretability of classical Shewhart charts.
- High-breakdown estimators (e.g., MAD) tolerate up to 50% contamination without limit distortion.
- Reduces false-alarm inflation caused by heavy-tailed process distributions or unscreened Phase I anomalies.
- Compatible with standard supplementary run rules and CUSUM/EWMA adaptations.
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 a robust control chart when Phase I reference data are suspected to contain outliers, special-cause events that were not screened out, or come from a process with heavy-tailed or skewed measurement distributions — situations where classical Shewhart charts would produce distorted control limits. It is particularly valuable in early production ramp-up, short-run manufacturing, chemical batch processing, and environmental monitoring where clean baseline data are difficult to obtain. Do not use it as a substitute for proper Phase I data cleaning: robust estimation compensates for a modest fraction of contamination (typically up to 25–50 % breakdown point), not for fundamentally wrong data collection. When process data are confirmed normal and outlier-free, classical charts are equally valid and simpler to explain.
Strengths & limitations
- Produces reliable control limits even when Phase I baseline data contain a moderate fraction of outliers or special-cause events.
- Inherits the real-time monitoring simplicity and interpretability of classical Shewhart charts.
- High-breakdown estimators (e.g., MAD) tolerate up to 50% contamination without limit distortion.
- Reduces false-alarm inflation caused by heavy-tailed process distributions or unscreened Phase I anomalies.
- Compatible with standard supplementary run rules and CUSUM/EWMA adaptations.
- Robust estimators are statistically less efficient than the mean and standard deviation under perfect normality, resulting in slightly wider limits from the same sample size when data are truly clean.
- The breakdown advantage applies primarily to Phase I limit setting; Phase II observations are still evaluated individually and a single outlier can still trigger a signal.
- Communication to process operators requires explanation of why limits differ from the familiar classical chart.
- Software support is less widespread than for classical Shewhart charts; practitioners may need to implement robust estimators manually.
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 is the breakdown point and why does it matter for control charts?
The breakdown point is the fraction of observations that can be arbitrarily corrupted before an estimator gives an arbitrarily bad result. The classical mean has a breakdown point of 1/n (a single outlier can move it without limit), while the median has a 50% breakdown point. For Phase I control limit estimation, a higher breakdown point means outliers in the reference data cannot silently distort the limits you will use for months of production monitoring.
Can I just delete outliers from Phase I data and use a classical chart instead?
Deleting confirmed special-cause events before setting limits is good practice and should always be done when the cause is identified. Robust charts are valuable when you cannot determine which Phase I points are genuine outliers versus rare but valid observations — or when the process made it impractical to collect clean reference data. They are complements to, not replacements for, Phase I data screening.
How do robust control charts relate to nonparametric control charts?
Nonparametric (distribution-free) charts make no distributional assumptions and use rank-based or sign-based statistics. Robust charts typically assume an underlying symmetric distribution but protect against contamination within that framework. Nonparametric charts offer broader distribution-free guarantees; robust charts are often easier to interpret and explain in terms of standard sigma limits.
Are there robust versions of CUSUM or EWMA charts?
Yes. Robust CUSUM and EWMA charts replace the classical mean and standard deviation in the recursive updating equations with resistant estimators, or use Huber-type M-estimators for the monitored statistic. These combine the small-shift detection sensitivity of CUSUM/EWMA with resistance to outliers in the incoming data stream.
What sample size do I need for Phase I robust estimation?
The same general guidance as classical charting applies — at least 20–25 subgroups of size 4–5 (or 100+ individual observations) to stabilize limit estimates. Robust estimators do not reduce the required Phase I sample size; they protect the estimates you do have from contamination.
Sources
- 1.Tatum, L. G. (1997). Robust estimation of the process standard deviation for control charts. Technometrics, 39(2), 127–141.
- 2.Rocke, D. M. (1989). Robust control charts. Technometrics, 31(2), 173–184.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Robust Control Chart. ScholarGate. https://scholargate.app/experimental-design/robust-control-chart