Control Chart

A control chart is a graph used in statistical process control (SPC) to study how a measured process behaves over time. You plot each measurement in time order and compare it against three reference lines the data itself defines — a center line at the process average, and an upper and a lower control limit, usually set three standard deviations above and below that average. Walter Shewhart devised the tool at Bell Labs in the 1920s, which is why it is also called a Shewhart chart. The control limits are not targets or specification limits chosen by a customer. They describe the range of variation the process produces when nothing unusual is happening, so a point inside them looks like ordinary noise and a point outside them looks like a genuine signal worth chasing.

Why bother distinguishing the two? Because every process varies, and reacting to ordinary variation as if it were a problem makes things worse. Shewhart's insight was to split variation into two kinds. Common-cause variation is the routine, built-in noise of a stable process, and it lives inside the control limits. Special-cause variation is something new — a broken machine, a changed supplier, a data-entry error — and it shows up as a point beyond a limit or as a non-random pattern, such as a long run on one side of the center line. A control chart tells you which kind you are looking at. That guidance keeps teams from tampering with a healthy process and prompts them to act only when the chart shows a real, assignable cause.

A control chart is easy to confuse with a run chart, and the difference is the control limits. A run chart is the simpler tool — just your measurements plotted in time order, often with a median line — and it helps you spot trends, shifts, and obvious oddities by eye. It has no statistical limits, so it cannot tell you whether a swing is meaningful or merely noise. A control chart adds exactly that: limits calculated from the process's own variation, plus formal rules for reading them. So a run chart shows the story of the data over time, while a control chart judges it against a statistical boundary. If you only need to watch a trend, a run chart is enough. If you need to decide whether to act, you want a control chart.

The comparison also clarifies what a control chart is not. It is not a specification chart. Specification limits come from what a customer or a design requires, and a process can sit comfortably inside its control limits yet still fail specification, or meet specification while drifting out of statistical control. The two answer different questions — control limits ask whether the process is stable and predictable, specification limits ask whether its output is acceptable. Confusing them is a classic error. A control chart is also distinct from a simple average or a pass-fail count, because it preserves the time order of the data and reads the pattern, not just the level. Its whole value is in watching a process move and catching the moment its behavior changes.

Use a control chart when you have a process you can measure repeatedly and want to keep stable — a call-center handle time, a fulfillment error rate, a daily ad spend, a conversion rate. Pick the right chart for your data, since different types suit counts, proportions, and continuous measurements. Gather enough baseline data before you fix the limits, because limits built on too few points, or on an already-unstable process, will mislead you. Then read the chart by agreed rules rather than gut feel: a point beyond a limit, a run of eight on one side, a clear trend. When a signal appears, hunt the special cause and remove it. When the chart is quiet, leave the process alone. Marketers increasingly borrow the method to monitor daily metrics without overreacting to every wobble.

The failures are predictable. People treat every up-and-down as a problem and tinker constantly, which adds variation instead of removing it — the tampering Shewhart warned against. They confuse control limits with specification limits and either panic or relax at the wrong moments. They set limits from a process that was never stable, so the chart is built on sand. They plot the wrong chart for the data, or too few points, and read noise as signal. And they let a chart go stale, never recalculating limits after a genuine process change. Used with discipline, a control chart is one of the plainest, most honest ways to see whether something is behaving — and to know, before you act, whether what you are seeing is real.

Control chart — the statistical process control tool Walter Shewhart devised at Bell Labs in the 1920s, plotting a process against a center line and control limits to separate ordinary variation from real signals.

Etymology: source.

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Areas of marketing where control chart is a core concern: