A three‑sigma limit (or ±3σ limit) sets control bounds at three standard deviations above and below a process mean. Under the empirical rule for a normal distribution, roughly 99.7% of observations lie within ±3σ of the mean, so points outside those bounds are unusually extreme and often indicate a special cause that should be investigated. In statistical process control, these limits are commonly used as the upper control limit (UCL) and lower control limit (LCL) on Shewhart (control) charts. (See Shewhart and the empirical rule.) [1][2]
Key concepts
– Mean (x̄): the average of your observations.
– Standard deviation (σ or s): a measure of variability or dispersion around the mean.
– ±3σ limits: UCL = mean + 3·σ ; LCL = mean − 3·σ.
– Empirical rule: for an approximately normal distribution ≈68% of values fall within ±1σ, ≈95% within ±2σ, ≈99.7% within ±3σ. [3][4]
How to calculate three‑sigma limits — step‑by‑step (practical)
1. Collect data correctly
• Gather a representative set of measurements from the process. Use consistent measurement methods and time periods. Consider subgrouping (e.g., samples of size n at regular intervals) for control charts.
2. Choose sigma estimator
• If you have the whole population, use population σ. For sample data, use the sample standard deviation s (denominator n−1) for an unbiased estimate. For control charts, sigma is often estimated from subgroup ranges or moving range methods (see Shewhart rules). [1][2]
3. Compute the mean (x̄)
• x̄ = (sum of all observations) / n.
4. Compute standard deviation
• Sample standard deviation s = sqrt( Σ(xi − x̄)² / (n − 1) ).
5. Compute three‑sigma limits
• UCL = x̄ + 3·s
• LCL = x̄ − 3·s
• If LCL < physical limit (e.g., negative when the measure cannot be negative), set LCL to that physical minimum.
6. Plot and interpret on control chart
• Plot new observations on the control chart with centerline = mean and control limits = UCL/LCL. Investigate special causes if:
• any point falls outside ±3σ, or
• non‑random patterns appear (runs, trends, or clustering), or
• multiple points violate ±2σ rules (Western Electric or Nelson rules). [1][5]
Worked example (10 test results)
Data: 8.4, 8.5, 9.1, 9.3, 9.4, 9.5, 9.7, 9.7, 9.9, 9.9
1. Mean: x̄ = 93.4 / 10 = 9.34
2. Sample standard deviation s ≈ 0.534 (calculated with n−1)
3. Three‑sigma band: 3·s ≈ 1.60
• UCL = 9.34 + 1.60 ≈ 10.94
• LCL = 9.34 − 1.60 ≈ 7.74
Interpretation: If future measurements fall outside ~7.74–10.94, that suggests the process may be out of control and warrants investigation. (If you used population σ instead, the numbers would be slightly different; be explicit about which estimator you use.) [1]
When to use three‑sigma limits
– Routine process monitoring in manufacturing and service operations where natural variability exists and occasional variation is acceptable.
– Setting statistical control limits on Shewhart control charts to detect special causes.
– Quick alerts for outliers in operations, finance, or quality assurance when a normal or approximately normal distribution of data is reasonable. [1][3]
Three sigma vs Six Sigma — main differences
– Three‑sigma: control limits at ±3σ from the mean (≈99.7% in a normal distribution). Good for detecting obvious special causes; widely used for control chart limits. [1][3]
– Six‑Sigma (methodology): a process improvement framework that aims for very low defect rates. Six sigma’s theoretical ±6σ corresponds to extremely small probability of defects; in practice Six Sigma often accounts for a possible 1.5σ long‑term shift and uses the 3.4 defects per million opportunities (DPMO) benchmark. Six Sigma is more prescriptive (tools, DMAIC, focus on capability) and targets much higher performance than a simple ±3σ control rule. [6][7]
How three‑sigma limits are used in practice
– Control charts (Shewhart charts): centerline = mean; UCL/LCL = ±3σ. Used to detect out‑of‑control conditions. [1]
– Process capability and monitoring: helps determine whether a process is stable before calculating capability indices (Cp, Cpk). Do not compute capability indices unless the process is in statistical control. [6]
– Early warning: triggers investigations into special causes (equipment, materials, operator practices).
– Continuous improvement: used as part of defect/rework reduction, root‑cause analysis, and corrective action cycles.
Limitations and cautions
– Assumes approximate normality. If data are highly skewed or discrete (counts, rates with low means), ±3σ from the mean may be misleading—use appropriate transformations or alternative charts (e.g., p‑chart, c‑chart, EWMA). [1][3]
– Sigma estimate matters: small samples produce noisy s estimates—use appropriate subgrouping or pooled estimators (e.g., average range) to estimate process sigma for control chart limits accurately.
– Control limits are not tolerance limits. ±3σ control limits describe expected process variability, not the customer specification. A process can be in control but not capable of meeting specifications.
– False alarms and missed signals: no single rule catches all real shifts; combine ±3σ with rules for runs/trends (Western Electric / Nelson rules) and with domain knowledge. [1][5]
Practical checklist to implement three‑sigma control limits
1. Define the metric and measurement procedure.
2. Collect a baseline dataset while the process operates under normal conditions (subgrouping if applicable).
3. Compute mean and appropriate sigma estimator (range, pooled s, or individual moving range).
4. Set UCL/LCL = mean ± 3·sigma and draw the control chart.
5. Use additional detection rules (e.g., runs, trends) to increase sensitivity to smaller shifts. [1][5]
6. When an out‑of‑control signal occurs, perform root‑cause analysis, document corrective actions, and only recalculate control limits after the process has been corrected and stabilized.
7. Periodically review and update the control chart and sigma estimates.
Further reading and sources
– Investopedia: “Three Sigma Limits” (Julie Bang) — explanation and context on control charts and the empirical rule. [1]
– National Library of Medicine / NIOSH: Walter A. Shewhart and the origins of control charts. / [2]
– Britannica: Standard deviation definition and interpretation. [3]
– University of North Dakota Libraries: Bell‑shaped curve and normal distribution. / [4] (see entries on bell-shaped curve / normal distribution)
– Six Sigma resources on principles and process capability (overview of Six Sigma methodology and capability indices). Example: “Six Sigma Principles” (SixSigma site) and Six Sigma primers showing the 1.5σ shift and DPMO concept. / [6]
– Practical guidance on control chart rules and detection (Western Electric / Nelson rules summaries). Example guidance: Dataforth application notes and operational guidance for control charts. [7]
The bottom line
Three‑sigma limits are a simple, widely used statistical tool to set control bounds and detect unusual variation in a process. They are quick to compute and effective when the data are approximately normal and the sigma estimate is reliable. Use them as part of a broader statistical process control program, combine them with other rules and root‑cause methods, and remember that they are monitoring tools—not substitutes for careful measurement practice, capability studies, or continuous improvement methodologies such as Six Sigma.