Gage R&R (Repeatability & Reproducibility)

Before you can trust any SPC data, you need to know whether your measurement system is capable. A Gage R&R study answers a simple question: when you see variation in your data, how much is real process variation and how much is measurement noise?

The AIAG MSA manual sets clear thresholds: if measurement system variation consumes less than 10% of total variation, the gage is acceptable. Between 10–30%, it may be acceptable depending on the application. Above 30%, the gage needs improvement before you can draw any reliable conclusions about the process.

The practical consequence is significant. If your gage consumes 40% of observed variation, your Cpk calculations are diluted by measurement noise. You might conclude a process is incapable when the real problem is the gage, not the process. Every SPC program should start with a Gage R&R study — yet many skip it.

EntropyStat's entropy-based analysis is particularly sensitive to measurement system quality because the EGDF works with the data as-is, without smoothing or averaging that might mask measurement noise.

When measurement variation is high relative to process variation, traditional methods can still produce "acceptable" results because the normal distribution assumption acts as a built-in smoother. The EGDF does not smooth away noise — it faithfully represents whatever the data shows, including measurement artifacts. This means a poor gage shows up immediately as unusual distribution shapes or unexpected ELDF clusters.

This is actually an advantage. By running EntropyStat on Gage R&R study data, you can detect measurement system problems that traditional %GRR calculations miss — such as operator-specific biases that create bimodal distributions, or non-linear gage behavior that produces skewed repeatability distributions. The ELDF's cluster detection can separate operator effects without requiring the balanced design that ANOVA-based Gage R&R demands.