Must a capability study always produce Cpk?

No. It may focus on common-cause variation, and other measures may suit the data or requirements better.

Can a stable process be incapable?

Yes. It can consistently exhibit too much variation or operate too close to an unacceptable limit.

Is there one universal sample size?

No. The needed evidence depends on the process, sampling structure, operating coverage and precision required for the decision.

AI Glossary / Statistics and probability / Intermediate

Capability study

Definition

A capability study investigates the common-cause variability of a process: the variation associated with its ordinary operating system rather than a specific unusual disturbance. In practical quality work, it often also compares a stable process's output with specification limits. The study is broader than calculating a capability index from a convenient batch of measurements.

What it means

The source glossary uses capability study for studying natural process variability. Comparing that variability with requirements is a common application, not the entire definition. A useful study establishes what process, measurement characteristic, operating conditions and time period the conclusion describes. It can reveal that a stable process needs improvement or that an unstable process first needs investigation. The word capability here concerns statistical process performance, not an AI system's general skills or an employee's competence.

How it works

Define the characteristic and, where relevant, its requirements. Check that the measurement system can resolve meaningful variation. Collect data in a way that represents the intended process, retaining time order and rational subgroup information when appropriate. Assess stability and investigate unusual changes before interpreting a single underlying distribution. Estimate the relevant variation and examine the distributional assumptions. If comparing with specifications, choose suitable capability measures, quantify estimation uncertainty and report observed performance separately from model-based expectations. Document what was excluded and why, and repeat the assessment after material process changes rather than assuming an old result remains current.

Three practical examples

Production qualification: study dimensional variation across representative operating periods before deciding whether a process can reliably meet requirements.
Measurement assessment: investigate repeatability and reproducibility to understand how much observed variation may come from the measuring system.
Improvement evaluation: repeat a comparable study after a justified process change to determine whether variation or centering has improved without changing the acceptance requirements.

Think of it this way

Think of testing how consistently a delivery system operates over ordinary working conditions, rather than timing one unusually quiet morning. A capability study asks what variation belongs to the system and how that variation relates to the task it must perform.

Worked example

Consider a simplified measurement model: observed value equals true process value plus independent, mean-zero measurement error. Suppose the observed standard deviation is 0.25 millimeters and a separate, valid measurement study establishes an error standard deviation of 0.07 millimeters. Under the stated independence model, observed variance is the sum of process and measurement-error variances. Estimated process variance is therefore 0.25 squared minus 0.07 squared = 0.0625 - 0.0049 = 0.0576, giving process standard deviation 0.24 millimeters. If a centered, stable normal process has limits 9 and 11 millimeters, the corresponding spread ratio is 2/(6 × 0.24), approximately 1.389. Using the observed spread directly gives 2/(6 × 0.25), approximately 1.333. This illustrative decomposition does not authorize subtracting arbitrary gauge estimates from every capability report. Independence, bias, uncertainty in both estimates, the reporting convention and the actual measurement objective must be considered. A negative variance difference would signal that this simple calculation cannot provide the assumed process-variance estimate.

Illustrative case study

Fictional case: Clearline Packaging starts a capability study after customers report inconsistent seal widths. The team samples across shifts and records material changes, machine settings and measurement operators. A time plot shows distinct shifts after adjustments, so analysts investigate those changes before publishing one capability number. They also check whether operators measure width consistently. Once the operating procedure is defined and assessed, the team reports the conditions to which the study applies and plans continued monitoring. This fictional case describes a structured investigation and reports no quantified quality improvement.

Common misconceptions and mistakes

A capability study is not the same as an acceptance-sampling plan for deciding whether to accept a particular lot. It is also not a proof that a process will never produce an unacceptable item. Statistical control and conformity to specifications are different questions. Pooling multiple machines or products can hide distinct distributions, while studying only ideal conditions can overstate ordinary performance. Removing inconvenient observations without an identified, documented reason can distort the result. Increasing sample size reduces some estimation uncertainty but cannot repair an unrepresentative sampling plan or an unsuitable measurement system.

Business use

Businesses can use capability studies to guide process development, supplier discussions and improvement priorities. Agree on the study's scope and criteria before collecting data, including whether the interest is short-term repeatability, routine production or broader operating variation. Preserve the raw measurements and their context. Use AI-assisted analysis to organize patterns or flag potential changes, while verifying the underlying measurements and statistical reasoning. Present findings with their limits so managers can distinguish an engineering improvement opportunity from a monitoring issue or a measurement problem.

Check your understanding

Q: Why are variances subtracted in the worked example rather than standard deviations?

Answer: A: Under the specified independent additive-error model, variances add. Subtracting 0.07 directly from 0.25 would produce 0.18, which does not follow from that model.

Related terms

common-cause variation
process stability
process capability index
measurement-system analysis
gauge repeatability and reproducibility
specification limit and acceptance sampling
evidence quality and possible operational uses

Learn next

design a sampling plan that captures the ordinary sources of variation in a familiar process. Compare a time plot with a pooled histogram, and identify what information pooling loses. Then examine how the study's conclusion changes when measurement variation and uncertainty are included explicitly.

Sources

Sources: NIST glossary, capability study and process capability study definitions: https://www.itl.nist.gov/div898/handbook/glossary.htm ; Minitab, data considerations for capability analysis: https://support.minitab.com/en-us/minitab/help-and-how-to/quality-and-process-improvement/capability-analysis/how-to/capability-analysis/normal-capability-analysis/before-you-start/data-considerations/ ; NIST, gauge studies and uncertainty components: https://www.itl.nist.gov/div898/handbook/mpc/section4/mpc46.htm .

Previous: Capability index, Cpk Next: Capacity Scaling Minimum Cost Flow