Attribute Agreement Analysis in Simple Terms
A Gauge R&R study works for measured values such as a diameter. Many checks are not measured: a visual inspector decides good or bad, or a go/no-go gauge fits or does not. Attribute agreement analysis is the equivalent study for those decisions and is part of MSA as described in the AIAG MSA manual.
How It Works
- Choose about 30 to 50 parts, including good, bad and borderline ones.
- Have an expert set the reference decision for each part.
- Two or three appraisers inspect each part at least twice, in random order, without seeing earlier results.
- Calculate agreement within each appraiser, between appraisers and against the reference, usually with percent agreement and the kappa statistic.
| Kappa value | Usual interpretation (AIAG) |
|---|---|
| 0.75 or more | Good to excellent agreement |
| 0.40 to 0.75 | Marginal |
| Below 0.40 | Poor agreement |
Example
A painted bumper line has three inspectors judging surface defects. In a 40-part study, one inspector agrees with the reference only 70 percent of the time and passes several parts with dirt inclusions. The team adds boundary samples and a lighting standard, retrains, and the repeat study shows kappa above 0.9 for all three.
Common Mistakes
- Using only obvious good and bad parts, which makes agreement look better than it is.
- Letting appraisers know which part they are looking at.
- Not checking against a reference, so a team can agree with each other and still be wrong.
The OperaHive calibration management software keeps Gauge R&R and measurement system analysis records with the gauge history.