What Is a Control Chart? SPC and the Nelson Rules Explained

Control charts are the heart of statistical process control — a simple, powerful way to tell ordinary variation from a real change. Here's the idea, minus the jargon.

Every process varies. The trick is telling the difference between the harmless wobble of a stable process and a genuine change you should act on. A control chart — invented by Walter Shewhart — does exactly that.

The core idea

Plot a measurement in time order. Draw a center line at the process average and control limits at three sigma above and below, estimated from the data's own short-term variation. As long as points fall randomly inside the limits, the process is in control — only common-cause noise. A point outside, or a non-random pattern, signals a special cause.

An annotated control chart with center line and control limits — Center line, control limits, and the zones the Nelson rules use.

Control limits are not spec limits

This trips people up constantly. Spec limits come from the customer or the requirement — what the output must be. Control limits come from the process — what it actually does. A process can be in perfect control yet fail spec (it's stable but off-target), or meet spec while drifting out of control.

The eight Nelson rules

A single point beyond three sigma is the obvious alarm, but small, sustained shifts hide in patterns. The eight Nelson rules catch them: a run of points on one side of center, a steady trend, too many points near the limits, oscillation, and so on. Together they make a control chart far more sensitive than just watching for the occasional outlier.

For slow drifts, go beyond Shewhart. CUSUM and EWMA charts accumulate small deviations, so they catch a gradual shift faster than a standard chart — useful when the change is subtle.

Ready to build one? See How to Build a Control Chart.

Download Stratum

Frequently asked questions

What is a control chart used for?

To monitor a process over time and tell ordinary, common-cause variation from a special-cause change — so you react to real signals and leave normal noise alone.

What's the difference between control limits and spec limits?

Control limits describe what the process actually does (calculated from its own variation, usually ±3 sigma). Spec limits describe what the output must be (set by the requirement). They are unrelated.

What do the Nelson rules detect?

Non-random patterns: points beyond 3 sigma, runs on one side of the center line, trends, oscillation, and points clustering near the limits — eight tests in all.

When should I use CUSUM or EWMA instead of a Shewhart chart?

When you need to catch small, slow shifts. CUSUM and EWMA accumulate minor deviations, so they react to a gradual drift sooner than a standard Shewhart chart.