Control Chart
A time-series chart with a center line and control limits that distinguishes common-cause variation from special-cause signals — the backbone of statistical process control.
// 01 — The chart
What it looks like
An X-bar control chart monitoring cereal fill weight. All points are within control limits (UCL/LCL), though sample 16 is approaching the upper limit and warrants attention.
// 02 — Definition
What is a control chart?
A control chart (also called a Shewhart chart, after its inventor) is a time-series plot designed to distinguish between common-cause variation (the natural, expected fluctuation in any process) and special-cause variation (unexpected signals that indicate something has changed).
Every control chart has three horizontal reference lines. The center line (CL) represents the process average. The upper control limit (UCL) and lower control limit (LCL) are typically set at ±3 standard deviations from the mean — not arbitrary thresholds, but statistically derived boundaries within which 99.73% of in-control data should fall.
When a point falls outside the control limits, or when a non-random pattern (runs, trends, oscillations) appears within them, the chart signals that the process is out of statistical control and investigation is needed. This makes the control chart both a monitoring tool and an early warning system.
Origin: The control chart was invented by Walter A. Shewhart at Bell Telephone Laboratories in 1924. His insight that variation is inherent in all processes — and that only special-cause variation should trigger action — laid the foundation for modern quality management and the Six Sigma movement.
// 03 — Anatomy
Parts of a control chart
// 04 — Usage
When to use it — and when not to
- You need to monitor a process over time and detect when something has changed
- Distinguishing random noise from real signals in manufacturing, healthcare, or service operations
- You want to set data-driven action thresholds rather than arbitrary targets
- Tracking KPIs like defect rates, cycle times, or response times sequentially
- You need to prove that a process improvement actually reduced variation
- Your data arrives in time-ordered samples and you suspect process instability
- Your data is not time-ordered — control charts require sequential observations
- You want to compare categories rather than track a single metric over time
- The process is known to be non-stationary by design (e.g., seasonal sales)
- You have fewer than 20–25 data points to establish reliable control limits
- You need to detect very small shifts quickly — consider CUSUM or EWMA charts instead
- Data is heavily autocorrelated — standard limits will produce many false alarms
// 05 — Reading guide
How to read a control chart
Follow these steps whenever you encounter a control chart in the wild.
Identify the three lines
Find the center line (process mean), UCL, and LCL. These establish the expected range of common-cause variation. The distance between UCL and LCL tells you how much natural spread the process has.
Check for out-of-control points
Look for any points above the UCL or below the LCL. These are immediate signals of special-cause variation — something outside the normal process has occurred and needs investigation.
Look for runs and trends
Even within limits, patterns can signal trouble. Seven or more consecutive points on one side of the center line (a run), or six or more points trending steadily up or down, suggest a systematic shift.
Check for hugging
If too many points cluster very close to the center line (within 1σ), the control limits may be too wide or the data may be stratified. This “hugging” pattern can mask real variation.
Investigate special causes
When a signal is detected, investigate its root cause before adjusting the process. Tampering with a stable process based on common-cause variation will actually increase variation — the opposite of the goal.
// 06 — Pitfalls
Common mistakes
Confusing control limits with specification limits
Control limits are calculated from the data and reflect what the process actually does. Specification limits are external requirements for what the process should do. They serve different purposes — never put spec limits on a control chart.
Recalculating limits after every new point
Control limits should be established from a stable baseline period and held constant. Continuously recalculating them absorbs special-cause variation into the limits, making real signals invisible.
Using the wrong chart type
An X-bar chart for individual measurements (use an I-MR chart), or a p-chart for continuous data (use an X-bar and R chart). The type of data determines which control chart variant is appropriate.
Ignoring non-random patterns
A point inside the limits can still be a signal if it is part of a run, trend, or oscillation. Apply the Western Electric rules or Nelson rules systematically — don’t just check for out-of-limit points.
Over-reacting to common-cause variation
Adjusting a stable process based on individual points within limits (tampering) increases variation. Only take action when a special cause is confirmed — otherwise, leave the process alone.
// 07 — In practice
Real-world examples
Manufacturing quality
Monitoring the diameter of machined parts every hour. An X-bar and R chart catches tool wear before defective parts are produced, reducing scrap and rework costs.
Healthcare safety
Tracking hospital-acquired infection rates monthly. A u-chart distinguishes normal fluctuation from a genuine outbreak, triggering targeted investigation only when warranted.
Software deployment
Monitoring API response times after each release. A control chart shows whether a new deployment changed system performance or whether observed variation is just noise.
Call center operations
Plotting average call handle time weekly. A control chart reveals whether a new training program genuinely reduced handle times or if the improvement is within normal variation.
// 08 — Quick reference
Key facts
| Also known as | Shewhart chart, SPC chart, process behavior chart |
| Invented | 1924 by Walter A. Shewhart at Bell Labs |
| Center line | Process mean (or median) |
| Control limits | ±3σ from the center line (99.73% coverage) |
| Key assumption | Data is independent and identically distributed when in control |
| Minimum data | 20–25 subgroups to establish reliable baseline limits |
| Signal rules | Western Electric rules, Nelson rules |
| Common fields | Manufacturing, healthcare, software, finance |
// 09 — Variations
Variations and extensions
X-bar and R chart
The classic pair: X-bar tracks sample means, R tracks sample ranges. Used for continuous data collected in subgroups of 2–10 observations.
p-chart / np-chart
For attribute (pass/fail) data. The p-chart plots proportion defective; the np-chart plots the count. Limits account for binomial variation.
CUSUM / EWMA chart
Advanced alternatives that accumulate information over time. Better at detecting small, sustained shifts in the process mean than standard Shewhart charts.
// 10 — FAQs
Frequently asked questions
What is a control chart?+
A control chart (also called a Shewhart chart, after its inventor) is a time-series plot designed to distinguish between common-cause variation (the natural, expected fluctuation in any process) and special-cause variation (unexpected signals that indicate something has changed).
When should you use a control chart?+
Use a control chart when you need to monitor a process over time and detect when something has changed. It also works well when distinguishing random noise from real signals in manufacturing, healthcare, or service operations, and when you want to set data-driven action thresholds rather than arbitrary targets.
When should you avoid a control chart?+
Avoid a control chart when your data is not time-ordered — control charts require sequential observations. It is also a poor fit when you want to compare categories rather than track a single metric over time, or when the process is known to be non-stationary by design (e.g., seasonal sales).
Is a control chart suitable for dashboards?+
Yes — a control chart can work well in dashboards as long as the panel is large enough for readers to perceive the encoded values, has a clear title, and includes the legend or axis labels needed to interpret it.
What category of chart is a control chart?+
Control Chart belongs to the Statistical family of charts. Charts in that family are designed to answer the same kind of question, so they often work as alternatives when one doesn't quite fit your data.
How do you read a control chart?+
Start with the axis labels and legend, then look at the overall shape before zooming into individual marks. Compare prominent features against the rest of the data, and verify any conclusion against the underlying numbers when precision matters.