A control chart is a time-ordered chart of a measure with a center line and upper and lower control limits. It shows whether a process is behaving as usual or whether something has changed, so a team reacts to real signals and leaves normal variation alone.
This guide covers the parts of a control chart, common and special cause variation, which chart type fits your data, the signals to look for, how to set one up and the mistakes to avoid. To build an individuals chart from your own readings, use the control chart calculator. The wider method is on the statistical process control page.
Key takeaways
Control limits come from the process's own data. They show what it does, not what it should do.
Common cause variation is the normal noise of a system. Special cause variation comes from something unusual.
Pick the chart by the data: measured values or counts, single readings or subgroups.
A point outside the limits is one signal. Runs, trends and cycles are others.
A stable process can still be a poor one. Stability first, then capability.
What is a control chart?
Walter Shewhart drew the first control chart in 1924 at Western Electric, in the engineering group that soon became Bell Telephone Laboratories. His 1931 book Economic Control of Quality of Manufactured Product set out the method. The idea is still the same: plot data in time order, work out how much it normally varies, and draw limits around that range.
Points inside the limits with no unusual pattern suggest the process is stable and predictable. A point outside, or a non-random pattern, suggests a special cause worth looking into. The chart does not judge whether a single result is good or bad. It tells you whether the process has changed.
Parts of a control chart
Every control chart has the same building blocks:
Data points
Measurements in the order they happened, such as hourly diameters, daily defects or weekly cycle time.
Center line
Usually the average of the data.
Upper and lower control limits
Usually three standard deviations either side of the center line, calculated from the data.
Time axis
When each reading was taken. Order matters.
Signal rules
The patterns that count as evidence of a change.
Control limits are not specification limits. Specifications come from the customer. A process can sit comfortably inside its control limits and still make parts outside specification.
Common cause and special cause variation
Telling the two apart is the whole point of the chart:
| Question | Common cause | Special cause |
|---|---|---|
| What it is | The normal, built-in variation of the system | Variation from something unusual or new |
| Example | Call handling times vary as customers ask different questions | A software release slows every order from Tuesday on |
| On the chart | Points scattered randomly inside the limits | A point outside the limits, or a run, trend or cycle |
| The right response | Change the system if the overall level is not good enough | Find out what changed at that time and deal with it |
What it is
The normal, built-in variation of the system
Special cause: Variation from something unusual or new
Example
Call handling times vary as customers ask different questions
Special cause: A software release slows every order from Tuesday on
On the chart
Points scattered randomly inside the limits
Special cause: A point outside the limits, or a run, trend or cycle
The right response
Change the system if the overall level is not good enough
Special cause: Find out what changed at that time and deal with it
Two mistakes are common. Treating common cause as special leads to tampering: adjusting a stable process in response to noise, which usually makes it worse. Treating special cause as common means missing a real problem because the latest number looked close enough.
Which control chart fits your data?
Choose by the type of data and how you collect it:
| Chart | Data | Example |
|---|---|---|
| I-MR (individuals and moving range) | One measured value at a time | Daily report completion time, one batch's viscosity |
| X-bar and R | Measured values in small subgroups, often 2 to 9 | Five shafts measured from each production run |
| X-bar and S | Measured values in larger subgroups | Ten or more readings per sample |
| p chart | Share of defective units, sample size may vary | Share of incomplete forms each week |
| np chart | Number of defective units, constant sample size | Defective units in a daily sample of 200 |
| c chart | Count of defects, constant area of opportunity | Surface defects per standard panel |
| u chart | Defects per unit, area of opportunity varies | Errors per invoice when invoice length varies |
I-MR (individuals and moving range)
One measured value at a time
Example: Daily report completion time, one batch's viscosity
X-bar and R
Measured values in small subgroups, often 2 to 9
Example: Five shafts measured from each production run
X-bar and S
Measured values in larger subgroups
Example: Ten or more readings per sample
p chart
Share of defective units, sample size may vary
Example: Share of incomplete forms each week
np chart
Number of defective units, constant sample size
Example: Defective units in a daily sample of 200
c chart
Count of defects, constant area of opportunity
Example: Surface defects per standard panel
u chart
Defects per unit, area of opportunity varies
Example: Errors per invoice when invoice length varies
Measured data, such as time, weight or length, detects small changes with fewer points than counts do. When in doubt with single readings, an I-MR chart is a sound start.
Signals to look for
A point beyond a limit is the clearest signal, but not the only one. Rule sets such as the Western Electric and Nelson rules also flag:
- A point beyond a control limit: Something unusual probably happened at that time.
- A run on one side: Eight or more points in a row above or below the center line, a sign the average has shifted.
- A trend: Six or more points steadily rising or falling, such as tool wear or a growing backlog.
- Points near a limit: Two of three in a row beyond two standard deviations on the same side.
- Cycles: Repeating waves that line up with shifts, days of the week or batches.
- A change in spread: Points suddenly scattering more or less widely.
Use a small, agreed set of rules. Applying many rules at once raises the chance of false alarms.
How to set up a control chart
- 1
Pick the process and measure. A repeated activity where variation matters, such as order accuracy, cycle time or a key dimension.
- 2
Write an operational definition. Exactly what is measured, how, how often and by whom.
- 3
Collect data in time order. About 20 to 25 points is a common starting point. Treat limits from fewer as provisional.
- 4
Calculate the center line and limits. Using the formulas for your chart type.
- 5
Plot and annotate. Add each new point in order and mark known changes, such as a new supplier or a staffing change.
- 6
Look for signals. Apply your agreed rules.
- 7
Investigate quickly. The closer the review is to the event, the easier it is to learn what happened.
- 8
Improve the system. If the process is stable but not good enough, change the process, then recalculate limits once the new level settles.
Control chart examples
Any repeated, measured work can be charted:
Manufacturing
Hourly diameter of a machined part, to see whether a machine is drifting.
Customer service
Daily average response time, to spot staffing or system problems.
Healthcare
Weekly patient wait time, to check whether a scheduling change helped.
Software
Escaped defects per release, to see whether a testing change reduced variation.
Finance
Invoice errors per batch, to catch training or system issues.
Projects
Cycle time of recurring work packages, to see whether a new approval step added delay.
Control charts in project management
Project managers, and anyone studying for the PMP exam, meet control charts as a quality tool. On a project they suit repeated work: review times, defect counts, change requests or task cycle times.
The rule that matters most is a consistent definition. If teams define 'done' or 'defect' differently, the chart shows measurement confusion, not process behavior. Used well, a chart lets a project manager show a sponsor whether a delay was a one-off or a sign that the process has changed.
Control chart vs run chart, histogram and Pareto chart
These charts look alike but answer different questions:
| Chart | Shows | Differs from a control chart |
|---|---|---|
| Run chart | Data over time with a median line | Has no calculated control limits |
| Histogram | The distribution of values | Loses the time order |
| Pareto chart | Categories ranked by frequency or impact | Ranks problems rather than monitoring stability |
| Pre-control chart | Color zones based on specification limits | Judges output against specifications, not process behavior |
Run chart
Data over time with a median line
Differs from a control chart: Has no calculated control limits
Histogram
The distribution of values
Differs from a control chart: Loses the time order
Pareto chart
Categories ranked by frequency or impact
Differs from a control chart: Ranks problems rather than monitoring stability
Pre-control chart
Color zones based on specification limits
Differs from a control chart: Judges output against specifications, not process behavior
A run chart is a good first step when you have little data. Add control limits once you have enough points.
Three questions for every chart review
Keep reviews short and focused on action:
Is the process stable?
Look for points outside the limits, runs, trends and cycles.
Is the performance good enough?
Compare the stable level with what customers need, using process capability for specifications.
What do we do next?
Investigate a special cause, redesign the system or keep monitoring.
Common control chart mistakes
Most mistakes are about interpretation, not math:
- 1
Too little data. Limits from a handful of points move a lot.
- 2
Mixing processes. Different products, sites or teams on one chart hide signals.
- 3
Treating limits as targets. They describe variation; they are not goals.
- 4
Confusing limits with specifications. A stable process can still miss requirements.
- 5
Reacting to every point. Adjusting a stable process to noise makes it worse.
- 6
Recalculating limits after every bad result. The limits stop meaning anything.
- 7
Changing the measurement midstream. Old and new data no longer compare.
- 8
Blaming individuals. Most variation comes from the system.
Frequently asked questions
What is a control chart used for?
How are control limits calculated?
What are the types of control charts?
What is the difference between control limits and specification limits?
Choosing SPC Software?
Read the buying guide: what to look for, mistakes to avoid and questions to ask vendors.
Related guides
- FMEA: failure mode and effects analysis, step by stepFMEA explained: the worksheet columns, eight steps, how to score severity, occurrence and detection, RPN vs action priority, and worked examples.Read the guide
- Kaizen event: meaning, steps and a 3-5 day agendaA kaizen event is a focused workshop that improves one process in 3-5 days. Learn the meaning, roles, agenda, steps, measures and common mistakes.Read the guide
- Process improvement methodologies: 9 methods comparedProcess improvement methodologies compared: PDCA, lean, kaizen, Six Sigma, TQM, theory of constraints, BPR and retrospectives, with how to choose and combine.Read the guide
- Root cause analysis example: 6 worked samples and a templateA root cause analysis example for six settings, from a machine stoppage to warehouse picking errors, plus root cause statement wording and common mistakes.Read the guide
Sources
Books and papers show their publication year. Web pages show the month we last checked them.
- 1.Walter A. Shewhart, Economic Control of Quality of Manufactured Product (D. Van Nostrand, 1931)
- 2.Western Electric Company, Statistical Quality Control Handbook (Western Electric, 1956)
- 3.Donald J. Wheeler, Understanding Variation: The Key to Managing Chaos (SPC Press, 1993)
- 4.Douglas C. Montgomery, Introduction to Statistical Quality Control, 8th edition (Wiley, 2019)