Is SPC only a control chart?

No. It includes data collection, interpretation, investigation, and improvement.

Does a signal identify the cause?

No. It indicates evidence that deserves investigation under the chosen rules.

Can an in-control process still perform poorly?

Yes. Stable performance may fail specifications or business needs.

AI Glossary / Statistics and probability / Intermediate

Statistical process control

Definition

Statistical process control, or SPC, uses statistical methods and process knowledge to monitor performance over time, detect meaningful changes, and support improvement. Control charts are a central tool, but SPC also includes disciplined data collection, investigation, and response. Its purpose is to distinguish routine process variation from evidence that the process has changed. SPC is not simply checking whether each output meets a specification, and it is not the automatic adjustment of equipment after every fluctuation.

What it means

A process can fluctuate even when its underlying operating conditions remain stable. SPC provides a structured way to compare current observations with that expected variation. A signal indicates that investigation may be warranted; it does not identify the cause by itself.

The method connects measurement to action. A dashboard that displays control limits but has no reliable data definitions, investigation process, or responsible owner is incomplete. Conversely, reacting to every ordinary up-and-down movement can introduce extra variation. The quality of the response matters as much as the calculation of the chart.

How it works

Choose a process characteristic tied to a meaningful operational outcome. Define units, sampling frequency, subgrouping, and the measurement method. Select a chart appropriate to the data: measurements, counts, proportions, or another structure. Establish a baseline and investigate unusual observations before treating it as a stable reference.

Calculate control limits using the chosen model and its assumptions. Monitor new observations in time order with prespecified signal rules. Investigate signals using process records, measurement checks, and knowledge of changes in materials, equipment, staffing, or environment. Record findings and corrective actions rather than deleting inconvenient points without explanation.

Review whether independence, subgroup size, and the assumed distribution remain suitable. When a verified improvement changes the process, establish a new justified baseline. Keep customer specifications separate from control limits and evaluate capability as an additional question. SPC can reveal instability, but stable performance may still require systematic redesign to meet requirements.

Three practical examples

1. A production line monitors the fraction of items failing inspection in successive samples and investigates unusual increases.

2. A service operation tracks a consistently defined processing-time measure while accounting for changes in case mix and dependence.

3. A laboratory follows reference-standard measurements to detect shifts in a measurement system before they affect routine reporting.

Think of it this way

Think of learning the normal rhythm of an engine and listening for changes. Ordinary vibration is expected, while a new pattern may signal a problem. Knowing the rhythm helps focus attention, but diagnosing and fixing the engine still requires investigation.

Worked example

Consider an original p-chart illustration. Suppose a reliable historical baseline gives defect probability p0 = 0.04, and each new subgroup contains 200 independent items with the same defect probability under the baseline model. The subgroup defect proportion has standard deviation sqrt[p0(1 - p0)/200] = sqrt(0.0384/200), approximately 0.013856.

Three-standard-deviation limits are 0.04 plus or minus 3 times 0.013856. The upper limit is approximately 0.08157. The lower calculation is approximately -0.00157, so the plotted lower limit is truncated to zero because a proportion cannot be negative.

One new subgroup contains 20 defective items, giving 20/200 = 0.10. This exceeds the upper limit and signals under the stated rule. Another contains 12 defectives, giving 0.06, which does not exceed either limit. That second result is not proof of satisfactory quality or permanent stability.

The limits are based on a stipulated established baseline, not estimated from these two subgroups. The simple binomial assumptions may fail if defects cluster by batch or inspection conditions change. If subgroup sizes vary, their standard errors and limits must reflect the actual sizes. Exact false-alarm probabilities also differ from a simple continuous-normal approximation because defect counts are discrete.

Illustrative case study

In a fictional assembly business, a p-chart signals after a supplier change. Supervisors initially blame an operator, but the investigation finds that several shifts using the same material lot experienced the increase.

The team checks the material, confirms the affected production, and documents the response. It keeps the established baseline while resolving the special cause rather than widening the limits to make the signal disappear. A later process improvement is evaluated separately before a new baseline is adopted. The chart supports disciplined investigation rather than replacing it.

Common misconceptions and mistakes

Control limits describe process behavior under a monitoring model; specifications describe requirements. Mixing them can cause both missed signals and unnecessary interventions. Do not treat points inside limits as proof that every relevant pattern is ordinary, especially when prespecified run or shift rules apply.

Autocorrelation, overdispersion, changing denominators, and measurement drift can undermine simple chart interpretations. Adding many rules changes false-alarm behavior. Recomputing limits after every new observation can hide a sustained shift. A signal also does not establish causality: process knowledge and follow-up evidence are needed. SPC is most effective when its assumptions and response procedures are maintained over time.

Business use

SPC helps organizations detect deterioration, evaluate improvements, and avoid costly overreaction to routine variation. It can support manufacturing, services, laboratories, and data-processing operations when measures and models fit the process. Assign ownership of each chart, define investigation steps, and retain records of baseline changes. Connect the monitoring program to customer outcomes rather than treating chart compliance as the final objective.

Check your understanding

A p-chart's subgroup size changes from 200 to 50 while the baseline defect probability stays the same. Should the same standard-error-based limits automatically be retained?

Answer: No. The smaller subgroup has a larger standard error, so appropriate limits depend on its size and the chosen chart method.

Related terms

Control chart
Common cause
Special cause
Rational subgroups
process stability
p-chart
rational subgroup
process capability
corrective action

Learn next

Learn how data type determines chart choice, then study subgrouping and dependence. Practice linking signals to a documented investigation process and keeping specifications separate from control limits.

Sources

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

NIST/SEMATECH, Proportions control charts: https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc332.htm

The primary handbook guidance was consulted. The p-chart arithmetic and fictional case are original teaching examples, not an established monitoring plan for a real process.

Previous: Statistical parity Next: Statistical quality control