Measure Phase
Measure establishes a trustworthy baseline. The exam tests operational definitions, data collection, measurement system analysis, process stability, capability, and sampling. The central rule is simple: do not calculate capability from data you do not trust.
Core rules
- Define each data field operationally: what is measured, how, by whom, when, and at what resolution.
- Build a data collection plan with data type, stratification, sampling frame, sample size, frequency, and recording method.
- Complete MSA before using data. For continuous data, Gage R&R below 10% is generally acceptable, 10-30% conditional, above 30% unacceptable; ndc should be at least 5. For attribute data, use attribute agreement analysis or kappa.
- Check stability before capability. A control chart shows whether the process is predictable; capability indices are meaningless for an unstable process.
- Cp compares specification width with 6 sigma within; Cpk accounts for centering. Pp and Ppk use overall variation. Cpk of 1.33 is a common minimum, higher for critical CTQs.
- Report DPMO or sigma level with the opportunity definition and a confidence interval where relevant.
Worked example
A CTQ has LSL 8, USL 12, process mean 10.5, and within-process sigma 0.5. Cpu = (12 - 10.5)/(3 x 0.5) = 1.00. Cpl = (10.5 - 8)/(3 x 0.5) = 1.67. Cpk = min(1.00, 1.67) = 1.00. The process is centered high and is only three sigma from the upper specification. If stable and normal, the upper tail is about 1,350 ppm, and Cpk is below the common 1.33 target. The team should investigate the mean shift before celebrating the lower-tail margin.
Traps that cost marks
- Calculating Cpk before MSA and stability.
- Using Cp when the process is off-center.
- Mixing within and overall sigma in one index.
- Taking an arbitrary sample that excludes shifts, lines, or customer segments.
- Ignoring measurement resolution that is too coarse for the specification.
- Treating DPMO as defective units rather than defects per opportunity.