In general we must focus on what we are actually trying to estimate and what actions we will take.
for gage R&R the critical estiamte is the measurement error, whcih is expressed by the standard deviation. This comes from the repeated measures of each part. Estimating a standard deviation from a small sample size can result in large deviations from the true standard deviation. Howver, the improvement in the individual standard deviation (whether estimated by the range or the SD) from n=2 repeats to n=3 repeats is negligable. BUT if we increase the number of these repeats the average standard deviation will be less effected by a few large estimates of any individual set of repeats. So, we are better off using more parts (n=30) and fewer reps (n=2). Same number of data points, same 'burden' on the operators, better accuracy in our estimates.
Now, how many operators? it depends on the system. if there is really no operator effect then a single operator is sufficient. (be careful, some CMM measurements are more effected by how the operator 'fixtures' the part than by the CMM itself) If there is operator effect, then I typically use two or three. The estimate of the statistical variation of these operators is bogus. (in statistical language they are fixed effects not random effects) what you are looking for is if there is a difference that matters so the range of difference is where you really look. I recommend this becuase of what we do with the information. If there is a substantial difference, we will retrain the operators to 'behave' lke the best one and then we can retest each individual operator.
for gage R&R the critical estiamte is the measurement error, whcih is expressed by the standard deviation. This comes from the repeated measures of each part. Estimating a standard deviation from a small sample size can result in large deviations from the true standard deviation. Howver, the improvement in the individual standard deviation (whether estimated by the range or the SD) from n=2 repeats to n=3 repeats is negligable. BUT if we increase the number of these repeats the average standard deviation will be less effected by a few large estimates of any individual set of repeats. So, we are better off using more parts (n=30) and fewer reps (n=2). Same number of data points, same 'burden' on the operators, better accuracy in our estimates.
Now, how many operators? it depends on the system. if there is really no operator effect then a single operator is sufficient. (be careful, some CMM measurements are more effected by how the operator 'fixtures' the part than by the CMM itself) If there is operator effect, then I typically use two or three. The estimate of the statistical variation of these operators is bogus. (in statistical language they are fixed effects not random effects) what you are looking for is if there is a difference that matters so the range of difference is where you really look. I recommend this becuase of what we do with the information. If there is a substantial difference, we will retrain the operators to 'behave' lke the best one and then we can retest each individual operator.