MSA answers a simple but critical question: how much confidence can you place in the data used to control and improve the process? AIAG treats measurement-data quality as a foundation for better manufacturing decisions. MSA is broader than Gage R&R alone.
Separates repeatability, reproducibility and part-to-part variation for continuous measurements. QAPP uses a crossed ANOVA model with interaction and reports VarComp, Study Variation, %Study Var, %Contribution, ndc and optional %Tolerance.
Analyse Gage R&R →Bias is systematic difference between the measurement result and a reference value. A system can be repeatable and still measure consistently high or low.
Analyse Bias →Checks whether bias remains constant across the measurement range. Small error at one size does not guarantee the same accuracy at the ends of the range.
Analyse Linearity →Tracks a check standard over time and looks for drift, centre shifts or precision changes relative to a baseline period.
Analyse Stability →For visual inspections or OK/NOK, defect A/B/C classifications. Checks within-appraiser repeatability, between-appraiser agreement and agreement with a known standard.
Analyse Attribute Agreement →| Question | Study | Example |
|---|---|---|
| How much variation comes from measurement? | Gage R&R | 3 appraisers measure the same parts repeatedly |
| Does the system measure systematically high/low? | Bias | Repeated readings of a 50.000 mm reference |
| Does bias change with size? | Linearity | References distributed from 10 to 50 mm |
| Does the system change over time? | Stability | Daily check-standard measurements |
| Do appraisers classify the same way? | Attribute Agreement | Visual OK/NOK inspection |
%Contribution uses variance components and shows how much total variance is associated with each source. %Study Var compares standard deviations and therefore does not sum to 100%. %Tolerance compares measurement-system variation with technical tolerance and is relevant when the measurement is used for specification decisions. ndc estimates how many distinct categories the system can distinguish relative to part variation.
Do not use a single metric as the verdict. Important appraiser×part interaction calls for investigation of geometry, method, fixturing, instructions and training. If the study parts cover too little of real process variation, %Study Var can misrepresent intended use.
A bias study needs an appropriate traceable reference. Linearity extends the question across the full operating range. NIST recommends more than four reference standards distributed over the range and repeated measurements at each level. In QAPP, the regression slope of bias versus reference shows whether systematic error changes with measured size.
Stability is the time dimension. An instrument may be calibrated today yet drift because of temperature, wear, environment, fixturing or other changes. QAPP uses a baseline to establish limits and then monitors the check standard with I-MR while keeping those limits fixed.
For nominal data, raw agreement is easy to understand but does not correct for agreement expected by chance. Kappa provides that adjustment. QAPP reports within-appraiser agreement, appraiser-versus-standard agreement, Fleiss κ between appraisers and a classification matrix. Minitab, citing AIAG MSA 4th edition, reports κ≥0.75 as good agreement and higher values such as 0.90 as preferred; final criteria still need to reflect classification risk and customer requirements.
QAPP uses public sources for formulas and explanations and AIAG for automotive methodological alignment. It does not reproduce the licensed AIAG manual and does not automatically certify compliance.