What Is Statistical Process Control?
Statistical Process Control (SPC) is a method for monitoring and controlling a process using statistical tools — primarily control charts — so that it operates at its full potential and produces conforming output as consistently as possible. Pioneered by Walter Shewhart at Bell Labs in the 1920s, SPC's central insight is that all processes vary, and the key to quality is distinguishing two kinds of variation.
Common cause variation is the natural, random "noise" inherent in a stable process. Special cause variation comes from a specific, assignable source — a worn tool, a new material lot, an untrained operator. A process showing only common cause variation is "in statistical control" and predictable. SPC's job is to flag special causes so you can act on real signals and avoid tampering with random noise.
Control Limits vs. Specification Limits
This distinction trips up many practitioners, so it is worth stating clearly:
- Control limits are the voice of the process. They are calculated from the data (typically the centerline ± 3 standard deviations of the plotted statistic) and describe what the process actually does.
- Specification limits are the voice of the customer. They come from the design drawing, contract, or standard and describe what is acceptable.
The two are completely independent. A process can be perfectly in control (no special causes) and still produce out-of-spec parts if its natural spread is wider than the tolerance. Never put specification limits on a control chart — they answer different questions.
Control Charts for Variable Data: X-bar and R
For continuous measurements (length, weight, diameter, temperature) collected in small subgroups, the workhorse is the paired X-bar and R chart:
- The X-bar chart tracks the average of each subgroup, monitoring shifts in the process center.
- The R chart tracks the range (max − min) of each subgroup, monitoring changes in process spread.
Control limits are computed using tabulated constants that depend on subgroup size n. For the X-bar chart, limits are X̄̄ ± A₂R̄; for the R chart, the limits are D₃R̄ and D₄R̄. Always interpret the R chart first — if the spread is unstable, the X-bar control limits (which depend on R̄) are not meaningful. For larger subgroups, X-bar and S (standard deviation) charts replace the range chart.
Control Charts for Attribute Data: p and c
When data is counted rather than measured, attribute charts apply:
| Chart | Tracks | Sample Size | Distribution |
|---|---|---|---|
| p chart | Proportion defective | Variable | Binomial |
| np chart | Number defective | Constant | Binomial |
| c chart | Defects per unit | Constant | Poisson |
| u chart | Defects per unit | Variable | Poisson |
The distinction between "defective" (a whole unit fails) and "defects" (the count of flaws, of which one unit may have several) determines whether you use the p/np family or the c/u family.
The Western Electric Rules
A single point outside the 3σ control limits is the most obvious signal, but patterns within the limits can also reveal special causes. The classic Western Electric rules divide each half of the chart into three zones (A = 2σ–3σ, B = 1σ–2σ, C = 0–1σ) and flag:
- Any single point beyond Zone A (outside the 3σ limit).
- Two of three consecutive points in Zone A or beyond (same side).
- Four of five consecutive points in Zone B or beyond (same side).
- Eight consecutive points on one side of the centerline.
Additional run rules flag trends (six points steadily increasing or decreasing) and other non-random patterns. Each rule is designed to catch a different kind of process upset while keeping false alarms low.
Process Capability: Cp and Cpk
Once a process is in statistical control, you can assess whether it is capable of meeting specifications. Two indices dominate:
Cp = (USL − LSL) ÷ 6σ — the ratio of the specification width to the process spread. Cp assumes the process is perfectly centered and measures potential capability.
Cpk = min[(USL − μ), (μ − LSL)] ÷ 3σ — accounts for how far off-center the process mean μ sits. Cpk is the actual capability and is always ≤ Cp. They are equal only when the process is perfectly centered.
A worked example: suppose USL = 110, LSL = 90, process mean μ = 102, and σ = 2.
- Cp = (110 − 90) ÷ (6 × 2) = 20 ÷ 12 = 1.67
- Cpk = min[(110 − 102), (102 − 90)] ÷ (3 × 2) = min[8, 12] ÷ 6 = 8 ÷ 6 = 1.33
The process could be very capable (Cp = 1.67), but because it runs off-center toward the upper limit, its actual capability drops to Cpk = 1.33. Common benchmarks: Cpk ≥ 1.33 (minimum capable, ~4σ), ≥ 1.67 (critical characteristics), and 2.0 (Six Sigma). A Cpk below 1.0 means the process is producing defects. You can compute these directly on the process capability calculator.
SPC in Practice
SPC is the natural home of the Control phase in a Six Sigma DMAIC project. To deploy it well: first bring the process into statistical control (eliminate special causes), then assess capability, and only then use the control chart for ongoing monitoring. Resist the urge to "tamper" — adjusting a stable process in response to common cause variation actually increases variation, a phenomenon Deming demonstrated with his famous funnel experiment.