Does an in-control process have no defects?

No. Its stable distribution may still produce unacceptable results.

Can a process be in control but off target?

Yes. Stability does not require the mean to equal the desired target.

Does one unusual point prove a special cause?

It provides a signal to investigate, subject to the monitoring model and its false-alarm behavior.

AI Glossary / Statistics and probability / Intermediate

Statistical control

Definition

Statistical control is the condition in which a process's observed variation is consistent with a stable system of common causes, rather than showing evidence of special causes that change its behavior. In the quality-control sense used here, an in-control process has a reasonably predictable distribution under the stated operating conditions. Statistical control does not mean zero variation, perfect products, or compliance with customer specifications. It describes stability, while capability concerns whether stable performance is suitable for the intended requirements.

What it means

Every process varies. Common causes are part of its current operating system, while special causes represent identifiable departures that may disturb that system. The distinction helps determine whether an individual signal deserves investigation or whether improvement requires changing the underlying process.

Evidence of statistical control is conditional on what was measured, when, and how. A chart with no signals does not prove that every aspect of the process is stable. A poorly chosen metric, short baseline, inappropriate limits, or ignored dependence can hide important changes. Stability must be assessed over a relevant range of operating conditions.

How it works

Choose a measurable process characteristic and a sampling plan that can reveal meaningful changes. Preserve time order and use rational subgroups when appropriate. Establish a baseline, investigate unusual patterns, and estimate suitable control limits using a method matched to the type of data and process.

Monitor subsequent observations against the reference and a specified set of signal rules. Points outside limits, sustained shifts, or other unusual patterns can prompt investigation. Such signals are evidence requiring interpretation, not automatic proof of a particular cause. Investigate measurement problems and operational changes as well as the process itself.

Do not repeatedly recalculate limits merely to absorb an unexplained deterioration. If a genuine process improvement creates a new stable regime, establish a justified new baseline. Keep control limits distinct from specification limits: the former describe expected process behavior under the monitoring model, while the latter state external requirements. Assess capability only with suitable stability evidence and distributional assumptions.

Three practical examples

1. A filling operation monitors fill-weight averages and within-subgroup variation to detect changes in its operating behavior.

2. A service team tracks a suitably modeled error rate over time and investigates a sustained shift after a workflow change.

3. A laboratory monitors reference measurements to distinguish routine measurement variation from an instrument problem requiring attention.

Think of it this way

Think of a familiar road whose bumps occur in a consistent pattern. The ride is predictable, but it may still be too rough for delicate cargo. Statistical control describes that predictability; improving the road enough for the cargo is a separate question about acceptable performance.

Worked example

Consider an original simplified monitoring model for individual measurements. Suppose a well-established stable normal process has known mean 100 and standard deviation 2. Three-standard-deviation reference limits are 100 minus 3 times 2 = 94 and 100 plus 3 times 2 = 106.

Now suppose the product specification requires individual values between 99 and 101. The process can remain stable around 100 while frequently producing values outside those much narrower specifications. Under the stipulated normal model, the conforming probability is P(-0.5 ≤ Z ≤ 0.5), approximately 0.3829, or 38.29%.

The corresponding capability index is Cp = (101 - 99)/(6 times 2) = 1/6, approximately 0.1667. Because the mean is centered, Cpk has the same value. This calculation demonstrates that statistical control and acceptable output are different properties.

A new individual value of 107 lies above the reference upper limit and warrants investigation under the stated monitoring rule. A value of 102 lies within the reference limits but outside the product specification. Neither conclusion depends on confusing the two sets of limits. The example assumes known parameters and an appropriate independent normal model; it is not a method for establishing limits from a few observations.

Illustrative case study

In a fictional packaging company, managers celebrate a control chart with no unusual signals. Customer complaints continue because the process's stable spread is wider than the permitted product range.

The quality team explains that daily operator adjustments are unlikely to solve the underlying variation and may make it worse. Engineers investigate equipment and material changes that could reduce the process spread. After an improvement, they verify a new stable baseline and reassess capability. The organization uses stability to guide improvement rather than treating it as the final quality objective.

Common misconceptions and mistakes

Do not read in control as good enough. A stable process can consistently miss a target or produce excessive defects. Conversely, a short run of conforming products can come from an unstable process.

Control limits are not confidence intervals for a population mean and are not interchangeable with specification limits. A point outside a limit does not reveal its cause by itself. Multiple signal rules increase opportunities for false alarms, and autocorrelation can invalidate simple chart interpretations. Avoid blaming operators for every routine fluctuation; meaningful improvement often requires changing common causes built into the process.

Business use

Statistical control helps businesses distinguish routine variation from evidence of change and choose appropriate responses. It supports more credible planning, improvement evaluation, and capability assessment. Establish ownership of signals, investigation steps, and baseline changes. Combine statistical evidence with process knowledge, and monitor metrics that matter to customers rather than relying on a convenient but incomplete proxy.

Check your understanding

A measurement is inside the process control limits but outside the customer's specification limits. Is it necessarily acceptable?

Answer: No. It may be consistent with routine process variation while still failing the customer requirement. The process may need systematic improvement.

Related terms

Common cause
Special cause
Control chart
Control limits
Specification limits
Rational subgroups
process stability
process capability
control limit
specification limit
rational subgroup

Learn next

Learn how control charts are constructed, then compare stability with capability. Study the effect of subgrouping, dependence, and signal rules before interpreting real process data.

Capability study

Sources

NIST/SEMATECH, glossary: https://www.itl.nist.gov/div898/handbook/glossary.htm

NIST/SEMATECH, Assessing Process Stability: https://www.itl.nist.gov/div898/handbook/ppc/section4/ppc45.htm

NIST/SEMATECH, What is Process Capability?: https://www.itl.nist.gov/div898/handbook/pmc/section1/pmc16.htm

The official conceptual guidance was read. The known-parameter normal calculation and fictional case are original teaching examples, not a completed process qualification.

Previous: Statistical confidentiality Next: Statistical Data Protection (SDP)