Measure
Data before analysis
Measure is about trustworthy data. Write operational definitions so two people measure identically, then build a data collection plan covering what, how, who, when, where, and how much. Stratify before collecting. Know the sampling options: simple random, stratified (guarantees representation of subgroups), systematic (watch for periodic patterns), and cluster (convenient but usually higher error).
Measurement systems
For continuous data, Gage R&R separates repeatability (equipment) from reproducibility (appraisers). Rough guidance: %GRR below 10% is acceptable, 10–30% is conditional on the application and cost of measurement, and above 30% is unacceptable. For attribute data use attribute agreement analysis with within- and between-appraiser kappa (≈0.7 acceptable, ≈0.9 good). Check resolution: the measurement increment should be about one-tenth of the tolerance or of the process variation, or the study cannot see real differences.
Baseline capability
Establish stability first, then capability. Cp and Pp describe spread only; Cpk and Ppk add centering. Pp/Ppk use overall sigma from the total spread; Cp/Cpk use within-subgroup sigma. Report the baseline as DPMO or sigma level.
Traps
Computing capability on an unstable process, treating ordinal ratings as continuous, and confusing %study variation with %tolerance when interpreting %GRR.
Rule of thumb: 30 or more points gives a rough capability estimate; 100 or more is preferred for Ppk.