Is a control chart the same as a run chart?

A run chart shows time order, while a control chart adds a statistical monitoring framework and limits.

Are three-sigma limits always appropriate?

No; chart choice, distribution, and monitoring objectives matter.

Can an in-control process be unacceptable?

Yes, because stability and meeting requirements are different questions.

AI Glossary / Statistics and probability / Intermediate

Control chart

Definition

A control chart is a statistical monitoring display that plots a process measure over time or successive samples against a reference center and control limits. It helps identify evidence of changes beyond the variation expected under a specified stable-process model. A signal prompts investigation; it does not by itself reveal the cause, prove a defect, or establish that a business requirement has been violated.

What it means

Processes vary even when nothing important has changed. A control chart provides a structured way to distinguish routine fluctuation from patterns that deserve attention. The chart combines time order, an appropriate measure, and decision rules rather than treating every movement as a meaningful improvement or deterioration.
Statistical control means that the process behaves consistently enough for the chosen monitoring model. It does not mean that the process meets customer requirements. A consistently slow service can be statistically stable, while an otherwise capable process can experience an unusual change that warrants investigation.

How it works

Define the process, observational unit, and measurement schedule. Choose a chart suitable for the data: subgroup means and variability, individual measurements, proportions, counts, or other justified statistics. Rational subgrouping should make the within-group variation and between-group changes relevant to the monitoring purpose.
Establish a defensible baseline and examine its stability before fixing prospective limits. Investigate special causes rather than automatically deleting inconvenient observations. Limits depend on the chosen statistic and sampling design; the standard deviation of an individual measurement is different from that of a subgroup mean.
Plot new observations in sequence and apply prespecified signal rules. A point beyond a limit is one possible signal. Runs or other nonrandom patterns can also matter, although adding rules changes the overall false-alarm behavior. Assign responsibility for responding and record the investigation and any corrective action.
Monitor variability as well as the center when relevant. Changes in sample size, autocorrelation, seasonality, mixture of populations, or measurement systems can invalidate simple chart assumptions. Adapt the method instead of assuming any time-series dashboard is a valid control chart.
Review limits after a documented process change and an appropriate new baseline. Continually recalculating them from every new observation can conceal the shift the chart is intended to detect. Keep specification limits and operational targets visually and conceptually distinct from statistical control limits.

Three practical examples

A manufacturer charts hourly subgroup mean dimensions alongside a chart of within-subgroup variation.
A service operation monitors a daily error proportion using limits that account for how many cases were reviewed.
An AI operations team monitors a suitably defined performance statistic and investigates signals alongside workload and measurement changes.

Think of it this way

Think of a control chart as a process's statistical listening device. It learns the ordinary background variation and highlights sounds that do not fit that pattern. It can tell the team to look more closely, but someone must still determine whether the cause is equipment, workload, measurement, or chance.

Worked example

Suppose a stable reference process has known mean 100 units and known individual standard deviation eight units. Every hour, analysts take a subgroup of sixteen independent measurements and plot the subgroup mean. Assume the normal model is appropriate and successive subgroups are independent.
The standard deviation of a subgroup mean is 8/sqrt(16) = 2 units. Three-standard-deviation limits for the mean chart are 100 - 3 x 2 = 94 and 100 + 3 x 2 = 106. An hourly mean of 107 is above the upper limit and triggers the specified investigation rule.
Using 100 plus or minus three times eight would instead give 76 and 124, which are inappropriate limits for these subgroup means. The distinction is between individual variation and variation of an average.
For this idealized normal model with known parameters and only the single-point rule, the probability that a stable-process mean lies beyond either limit is about 0.0027 per subgroup. Across repeated monitoring, false signals remain possible. Estimated limits, additional rules, dependence, or nonnormal subgroup means alter that behavior. The example illustrates chart construction, not proof of the reason for the observed 107.

Illustrative case study

Fictional case: Brookside Packaging plots average fill weight for regular production subgroups. An unusually high subgroup mean produces a signal. The team checks the weighing system, recent adjustments, and production records before changing the equipment.
It finds that a temporary setup change coincided with the signal and verifies the process after correction. The team keeps the original event and investigation in its records rather than erasing the point. It also examines the variability chart, because acceptable average fill weight can hide excessive variation among packages. Customer specifications remain a separate assessment from the control chart's stability question.

Common misconceptions and mistakes

Do not replace control limits with customer tolerances and call the result a statistical control chart. Do not assume every signal has a proven assignable cause, or that all points inside limits establish stability. Using the wrong denominator, pooling unlike processes, ignoring changing sample sizes, and treating correlated observations as independent can produce misleading alarms or hide real changes. Adjusting the process after every routine fluctuation can make variation worse.

Business use

Control charts help teams respond consistently to process evidence and reduce unstructured reactions to dashboards. Define who investigates signals, what contextual information is collected, and how a confirmed change is handled. Assess the cost of false alarms and missed changes when selecting the chart and its rules. For AI systems, chart performance only when evaluation data and outcome definitions remain meaningful over time.

Check your understanding

What standard deviation belongs in the worked mean-chart limits?

Answer: Two units, the standard deviation of subgroup means.

Does 107 prove which machine part failed?

Answer: No.

Should a chart of the center replace monitoring variability?

Answer: No; changes in spread can matter even when the center appears stable.

Related terms

Statistical process control
Control limits
Rational subgroups
common-cause variation
special-cause variation
rational subgroup
process capability
run chart
average run length

Learn next

Compare charts for individual values, subgroup averages, and proportions. Then study baseline assessment, false-alarm rates, and response procedures before introducing additional pattern rules or monitoring many measures simultaneously.

Sources

NIST/SEMATECH e-Handbook, What are Control Charts?: https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc31.htm NIST/SEMATECH e-Handbook, Shewhart X-bar and R and S Control Charts: https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc321.htm

Previous: Control break Next: Control factor