The Control Chart: Your Early Warning System for Process Problems
Every process produces variation. Output weights fluctuate, cycle times shift, defect counts move up and down from day to day. The question that separates a process owner from a fire-fighter is this: is what I am seeing normal noise, or is something actually wrong? A control chart answers that question with statistical rigour, replacing gut feel with a principled decision rule.
A control chart is a time-series plot — data in the order it was produced — overlaid with three reference lines: the process average in the centre, and an upper control limit (UCL) and a lower control limit (LCL) on either side. The limits are not specification limits set by a customer or manager; they are calculated from the data itself, typically placed at ±3 standard deviations from the mean. Any point inside the limits is behaving the way the process normally behaves. Any point outside them is a statistical signal that something has changed.
Common cause versus special cause variation
Walter Shewhart, who invented the control chart in the 1920s, drew a fundamental distinction between two types of variation. Common cause variation (sometimes called noise or chance cause) is the random, inherent variation built into every process. It produces a stable, predictable pattern over time. Special cause variation (also called assignable cause) is variation that comes from a specific, identifiable source — a new supplier batch, a machine that needs calibration, an operator who was trained differently, a seasonal spike in demand.
The practical consequence is enormous. If you react to common cause variation as if it were a special cause — tinkering with the process every time a point moves up or down — you actually increase variation. W. Edwards Deming called this tampering. Conversely, if a special cause appears and you dismiss it as normal noise, the underlying problem compounds. The control chart tells you which response is appropriate.
The eight Western Electric detection rules
A single point beyond the control limits is the most obvious signal, but the Western Electric Statistical Quality Control Handbook (1956) established eight patterns that also indicate a process has changed, even when every point stays within the limits:
One point beyond ±3σ. The classic out-of-control signal.
Two of three consecutive points beyond ±2σ on the same side. A cluster near the edge of the limits.
Four of five consecutive points beyond ±1σ on the same side. Sustained shift away from the centre.
Eight consecutive points on the same side of the centreline. A definitive shift in the process mean.
Six consecutive points steadily increasing or decreasing. A trend — often a tool wearing, a reagent degrading, or a gradual calibration drift.
Fifteen consecutive points within ±1σ. Appears orderly but usually means stratified sampling — two sub-populations mixed together.
Fourteen consecutive points alternating up and down. Systematic over-adjustment or a saw-tooth feeding pattern.
Eight consecutive points beyond ±1σ on both sides with none near the centreline. Two distinct populations are being charted together; disaggregate the data.
How to respond
When a special cause signal appears, stop and investigate before the evidence disappears. Document what was happening at that time: which shift, which raw material lot, which operator, which machine setting. Root-cause analysis at that moment yields far more than a retrospective hunt through logs a week later. Once the cause is identified, eliminate it to prevent recurrence.
When the process is in statistical control — only common causes present — the right response is to improve the process itself. Kaizen events, design of experiments, and process redesign are the tools for reducing common cause variation. Reacting to individual points will not help and will likely hurt.
Choosing the right chart type
The choice of chart depends on what you are measuring and how you collect the data. The individuals chart (I-MR) is the most versatile: use it when you have one measurement per subgroup, which is common in low-volume or slow processes. The X-bar and R chart is suited to subgroups of two to ten items collected at regular intervals — a classic in manufacturing. The p-chart tracks the proportion of defective items in variable-sized subgroups; the c-chart counts the number of defects per unit of constant size. Selecting the wrong chart type produces incorrect control limits and unreliable signals, so matching chart to data structure matters.
Getting started without specialised software
You do not need a dedicated statistical package to run a control chart. Excel handles individuals charts well: calculate the moving range between consecutive data points, find the average moving range (MR-bar), and set UCL = X-bar + 2.66 × MR-bar and LCL = X-bar − 2.66 × MR-bar. Free tools such as QI Macros (trial version) and the R language's qcc package produce professional charts in minutes. The investment is modest; the payback — distinguishing signal from noise on your first significant problem — is immediate.
If your organisation is ready to build a sustained Lean Six Sigma capability rather than fighting fires one chart at a time, XNM's strategic advisory practice works with leadership teams to embed data-driven improvement across the enterprise.