How Many Groups to Measure Accuracy?

[denton]

Theoretically, any system that accounts for the distance of all points from some reference, contains all the global variation information that is available. AGR contains all that information, but structures it awkwardly. In theory, it is better than extreme spread.

For five-shot groups the information content of range ("extreme spread") is 90% of theoretical maximum, so the practical difference is not worth the extra effort.

In effect, extreme spread contains practically all the available information, when n is small.

By the time you get to n=10, the effectiveness of extreme spread is down to 60% or so. That is the point where most statisticians kick over to using standard deviation, which has 100% information content.